<?xml version="1.0" encoding="UTF-8"?>
<!-- GENERATED by build_site.py — do not hand-edit. -->
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:content="http://purl.org/rss/1.0/modules/content/">
<channel>
  <title>Aplite blog</title>
  <link>https://aplite.ai/blog</link>
  <description>Notes on building and running Aplite.</description>
  <language>en</language>
  <lastBuildDate>Wed, 04 Nov 2026 00:00:00 +0000</lastBuildDate>
  <atom:link href="https://aplite.ai/blog/feed.xml" rel="self" type="application/rss+xml"/>
  <item>
    <title>What is OpenClaw, in simple terms (and how to try it without installing anything)</title>
    <link>https://aplite.ai/blog/what-is-openclaw</link>
    <guid isPermaLink="true">https://aplite.ai/blog/what-is-openclaw</guid>
    <pubDate>Tue, 20 Oct 2026 00:00:00 +0000</pubDate>
    <description>OpenClaw is an open-source AI assistant you run yourself. Gateway, agents, sessions, skills and channels explained plainly, what it does and does not do, and two ways to try it.</description>
    <content:encoded><![CDATA[<figure class="fig">
<a class="zoom" href="https://aplite.ai/assets/blog/what-is-openclaw.webp?v=abbc4fff" target="_blank" rel="noopener"><img src="https://aplite.ai/assets/blog/what-is-openclaw.webp?v=abbc4fff" alt="One gateway on a machine you control: chat apps on one side, a model and tools on the other." width="1560" height="820" decoding="async"></a>
<figcaption>One gateway on a machine you control: chat apps on one side, a model and tools on the other.</figcaption>
</figure>
<p>Search for <em>what is openclaw</em> and the first page answers three different questions.
Wikipedia tells the story: an Austrian developer, four names in two months, a star count.
The official docs explain the architecture: trust boundaries, runtimes, a WebSocket
protocol. Hosting companies explain it on a page that ends at their control panel. None
of them says in plain words what the thing is, what the five words in every tutorial
mean, and what it will and will not do. This page does, with the terms as the
<a href="https://docs.openclaw.ai">OpenClaw docs</a> use them, read on 20 October 2026.</p>
<h2 id="the-short-version">The short version</h2>
<p>OpenClaw is an open-source program that you run on a computer you control, and that turns
an AI model into an assistant you message from the chat apps you already use. The docs&rsquo;
own line: &ldquo;an open-source AI assistant that runs on your own hardware and meets you in
every chat app you already use.&rdquo;</p>
<p><em>Open source</em>: the code is on <a href="https://github.com/openclaw/openclaw">GitHub</a> under the
MIT license, about 393,000 stars on 20 October 2026, stewarded by the OpenClaw
Foundation, a US non-profit that sells nothing (&ldquo;No subscription. No hosted tier. No
token.&rdquo; on <a href="https://openclaw.ai">openclaw.ai</a>). <em>You run it</em>: on a laptop, a Mac mini or
a rented server, and only while that machine is on. <em>An AI model</em>: OpenClaw contains
none. It connects to one, from Anthropic, OpenAI, Google and others through a key or a
subscription login, or a local model through Ollama. The model thinks; OpenClaw is
everything around it: the chat connections, the memory on disk, the tools, the schedule.
The <a href="https://github.com/openclaw/openclaw/releases">current release</a> is 2026.10.10
(4 November 2026).</p>
<h2 id="the-five-words">The five words</h2>
<p><strong>Gateway.</strong> The one process that runs all the time: it holds the connections to your
chat apps, serves the browser dashboard, runs the scheduler and talks to the model, on
port 18789 of the machine itself by default
(<a href="https://docs.openclaw.ai/concepts/architecture">architecture</a>). When it is down,
nothing answers; a message sent while your laptop sleeps waits until it wakes. That is
why &ldquo;where does the gateway run&rdquo; is the first practical question.</p>
<p><strong>Agent.</strong> A configured assistant with its own folder, default model and history. The
folder, the workspace, holds plain markdown you can edit: <code>AGENTS.md</code> for operating
instructions, <code>SOUL.md</code> for persona, <code>USER.md</code> for who you are, <code>MEMORY.md</code> for long-term
memory (<a href="https://docs.openclaw.ai/concepts/agent">agent</a>). A fresh install has one agent,
<code>main</code>; more can be added, with different chats routed to each
(<a href="https://docs.openclaw.ai/concepts/multi-agent">multi-agent</a>).</p>
<p><strong>Session.</strong> A conversation. By default every direct message from every channel lands in
one rolling conversation, the main session, so what you said in Telegram is known when
you continue in the browser; groups get their own, and scheduled jobs start fresh ones
(<a href="https://docs.openclaw.ai/concepts/session">sessions</a>).</p>
<p><strong>Skills.</strong> Instructions for a specific job: a folder with a <code>SKILL.md</code> that names the
skill and says when and how to use which tools
(<a href="https://docs.openclaw.ai/tools/skills">skills</a>). Some ship with OpenClaw, some you
write, some you install from <a href="https://clawhub.ai">ClawHub</a>, the public registry, with
<code>openclaw skills install @owner/&lt;slug&gt;</code>; the docs call a third-party skill untrusted
code, to be read first.</p>
<p><strong>Channels.</strong> Where messages come in and go out. Telegram and the browser&rsquo;s WebChat are in
the core install; Discord, Slack, WhatsApp, Signal, iMessage, Matrix, Microsoft Teams,
Google Chat and a dozen more are official plugins
(<a href="https://docs.openclaw.ai/channels">channels</a>). An unknown sender gets a pairing code
instead of an answer until you approve them.</p>
<p>Four more, briefly: <strong>tools</strong>, what the agent can do (<code>read</code>, <code>write</code>, <code>exec</code>,
<code>browser</code>, <code>web_search</code>; <a href="https://docs.openclaw.ai/tools">tools</a>); <strong>memory</strong>,
<code>MEMORY.md</code> plus dated notes in <code>memory/</code>, so it remembers only what it wrote down
(<a href="https://docs.openclaw.ai/concepts/memory">memory</a>); <strong>cron</strong> jobs, a prompt on a
schedule delivered to a channel (<a href="https://docs.openclaw.ai/cli/cron">automations</a>); the
<strong>heartbeat</strong>, a check every 30 minutes by default, silent when nothing needs attention
(<a href="https://docs.openclaw.ai/gateway/heartbeat">heartbeat</a>).</p>
<h2 id="what-it-does-and-what-it-does-not">What it does, and what it does not</h2>
<p>It answers in your chat apps; reads and writes files in its workspace; runs commands on
its machine; opens a browser; searches the web; remembers across days in files you can
read; messages you first, on a schedule or when a heartbeat finds something; runs several
agents and sub-agents. The docs&rsquo; own examples: inbox and calendar briefings, research,
reminders, browser automation, a task sent from your phone
(<a href="https://docs.openclaw.ai/help/faq/what-is-openclaw">FAQ</a>).</p>
<p>What it does not do, from the same sources:</p>
<ul>
<li><strong>It has no model of its own.</strong> You bring a provider key, a ChatGPT or Claude
  subscription login, or a local model; the model is where the bill comes from. Every
  path, including the free ones: <a href="https://aplite.ai/blog/openclaw-without-api-key">OpenClaw without an API key</a>.</li>
<li><strong>It does not run while its machine is off.</strong> Sleep, network drops and OS updates stop
  the gateway; the upstream FAQ says to use a VPS after disconnects
  (<a href="https://docs.openclaw.ai/help/faq-first-run/providers-and-hosting">providers and hosting</a>).
  The free ways of keeping it on: <a href="https://aplite.ai/blog/openclaw-hosting-for-free">OpenClaw hosting for free</a>.</li>
<li><strong>It is not locked down by default.</strong> Sandboxing and exec approvals are off until you
  turn them on, and one gateway is one trust domain: &ldquo;a trusted single-operator
  assistant&rdquo; (<a href="https://docs.openclaw.ai/start/why-openclaw">why OpenClaw</a>,
  <a href="https://docs.openclaw.ai/gateway/security">security</a>).</li>
<li><strong>It is not a coding IDE.</strong> The docs say Claude Code or Codex is better for fast coding
  inside a repository; OpenClaw is the assistant around them, and can run both as its
  agent runtime
  (<a href="https://docs.openclaw.ai/concepts/agent-runtimes">agent runtimes</a>).</li>
<li><strong>It is not a hosted service.</strong> Hosting is what third parties sell,
  <a href="https://aplite.ai/blog/managed-openclaw-hosting-compared">seventeen of them</a> in October 2026; the
  hosted path below is one.</li>
</ul>
<h2 id="how-it-differs-from-a-chat-with-a-model">How it differs from a chat with a model</h2>
<p>ChatGPT, Claude.ai and Gemini are a model behind a website: you go to it, the
conversation lives there, and it touches what the vendor connected. OpenClaw inverts
that: the assistant comes to the messenger you already have open, its memory is files on
your disk, its hands are tools on your machine, and it acts on a schedule without being
asked, which no chat tab does. The model is a setting: <code>/new &lt;model&gt;</code> switches it
mid-conversation, and an agent can fail over to a second provider. The price is that you
operate it: machine, keys, updates and security are yours.</p>
<p>Beside it in the same search: <a href="https://github.com/NousResearch/hermes-agent">Hermes Agent</a>
by Nous Research, the same shape, self-hosted and messaging-first; and Claude Code and
Codex, coding harnesses OpenClaw can drive rather than replace.</p>
<h2 id="how-to-try-it">How to try it</h2>
<h3 id="install-it">Install it</h3>
<p>About five minutes, by the upstream
<a href="https://docs.openclaw.ai/start/getting-started">getting started</a> page.</p>
<ol>
<li>
<p><strong>Have Node 24.16 or newer, or 26.1 or newer</strong> (the installer adds it when missing)
   and a model: a provider key, a Claude Code or Codex login, or Ollama.</p>
</li>
<li>
<p><strong>Install.</strong> macOS, Linux or WSL2:</p>
<pre><code>curl -fsSL https://openclaw.ai/install.sh | bash
</code></pre>
<p>Windows, in PowerShell:</p>
<pre><code>iwr -useb https://openclaw.ai/install.ps1 | iex
</code></pre>
<p>Desktop apps: <a href="https://docs.openclaw.ai/install">install page</a>.</p>
</li>
<li>
<p><strong>Onboard.</strong> The installer opens the wizard; <em>Quick start</em> finds your login or key,
   verifies it with a real completion and opens the dashboard.</p>
</li>
<li>
<p><strong>Keep it running.</strong> Press Ctrl+C in the foreground gateway, install it as a
   service, then check it:</p>
<pre><code>openclaw gateway install
openclaw gateway status
</code></pre>
<p>The second command should show the gateway listening on port 18789.</p>
</li>
<li>
<p><strong>Talk to it.</strong> <code>openclaw dashboard</code> opens the Control UI; type a message. From your
   phone, Telegram is the fastest channel:
   <a href="https://aplite.ai/blog/openclaw-telegram-setup">OpenClaw in Telegram in five minutes</a>.</p>
</li>
</ol>
<p>If something fails, <code>openclaw doctor</code> reads the findings.</p>
<h3 id="without-installing-anything">Without installing anything</h3>
<p><a href="https://aplite.ai/docs/openclaw">Aplite</a> runs a standard OpenClaw instance for you: one gateway in its own
container on our server, currently 2026.10.10, set up from the browser. Create an account
with email or Google, no card on the Free plan; pick an agent, press Deploy; under a
minute later the workspace opens and you ask. No key to paste: models come through one
metered account, billed in credits, 1 credit = 1¢, token usage under each reply. On top
sit a markdown workspace the agent reads and writes, an approval queue for its changes,
and your phone: Telegram from a bot token, WhatsApp and WeChat by QR. What a managed instance gives up, the shell, raw
<code>openclaw.json</code>, local models and day-one releases, is on the
<a href="https://aplite.ai/docs/openclaw">OpenClaw page</a>; other hosts are in
<a href="https://aplite.ai/blog/managed-openclaw-hosting-compared">Managed OpenClaw hosting compared</a>. Aplite is
independent and not affiliated with the OpenClaw project.</p>
<h2 id="faq">FAQ</h2>
<h3 id="what-is-openclaw">What is OpenClaw?</h3>
<p>An open-source AI assistant you run on your own computer or server: it connects a model
of your choice to your chat apps, gives it tools such as files, shell, browser and web
search, keeps its memory in markdown files, and can act on a schedule.</p>
<h3 id="is-openclaw-free">Is OpenClaw free?</h3>
<p>The software is free and the project sells nothing. Running it costs a machine that stays
on and the model, billed per token unless it runs locally
(<a href="https://aplite.ai/blog/openclaw-hosting-for-free">OpenClaw hosting for free</a>).</p>
<h3 id="do-i-need-an-api-key-to-use-openclaw">Do I need an API key to use OpenClaw?</h3>
<p>Self-hosted, yes, or a substitute: a ChatGPT or Claude subscription login that onboarding
reuses, or a local model through Ollama. On a managed host that holds the model account,
Aplite among them, there is no key.</p>
<h3 id="how-does-openclaw-work">How does OpenClaw work?</h3>
<p>The gateway stays running. A message from a chat app goes to an agent with its
instructions, memory and skills; the agent calls the model, runs the tools the answer
needs, and the reply returns through the same channel. Scheduled jobs and the heartbeat
start the same loop without a message.</p>
<h3 id="who-makes-openclaw-and-is-it-owned-by-openai">Who makes OpenClaw, and is it owned by OpenAI?</h3>
<p>Peter Steinberger published it in late 2025 as Clawdbot, renamed Moltbot and then
OpenClaw in January 2026 (<a href="https://en.wikipedia.org/wiki/OpenClaw">Wikipedia</a>); the
OpenClaw Foundation, an independent non-profit funded by donations, stewards it since he
joined OpenAI in February 2026. OpenAI is one donor; the docs say donors do not control
the project and no lab&rsquo;s model is privileged in the code.</p>]]></content:encoded>
  </item>
  <item>
    <title>What OpenClaw really costs per month: server, models and the hidden lines</title>
    <link>https://aplite.ai/blog/openclaw-cost-per-month</link>
    <guid isPermaLink="true">https://aplite.ai/blog/openclaw-cost-per-month</guid>
    <pubDate>Tue, 20 Oct 2026 00:00:00 +0000</pubDate>
    <description>The OpenClaw bill is three lines: a server, the model tokens and your time. Seven ways to run it priced from primary sources on 20 October 2026, from a Mac mini with Ollama to a $20 managed plan, with the model line worked out per message and per day, and the heartbeat that quietly dominates it.</description>
    <content:encoded><![CDATA[<figure class="fig">
<a class="zoom" href="https://aplite.ai/assets/blog/openclaw-cost-per-month.webp?v=26f7b290" target="_blank" rel="noopener"><img src="https://aplite.ai/assets/blog/openclaw-cost-per-month.webp?v=26f7b290" alt="One month of OpenClaw as a receipt: the server line, the model line, and the lines nobody prints." width="1560" height="820" decoding="async"></a>
<figcaption>One month of OpenClaw as a receipt: the server line, the model line, and the lines nobody prints.</figcaption>
</figure>
<p>Ask a search engine how much OpenClaw costs per month and the answers span three
orders of magnitude: Hostinger&rsquo;s guide says
<a href="https://www.hostinger.com/tutorials/openclaw-costs">&ldquo;approximately $6 to $200+ per month&rdquo;</a>,
Kilo&rsquo;s says <a href="https://kilo.ai/articles/how-much-does-openclaw-cost">$0 to $520 and up</a>.
Ask an AI assistant for the cheapest way to run it around the clock and you get a server
price with a footnote that &ldquo;API usage is billed separately&rdquo;. Both are right, and neither
is the number you will pay. The software itself is free: OpenClaw is MIT-licensed, and its
site says <a href="https://openclaw.ai/">&ldquo;No subscription. No hosted tier. No token.&rdquo;</a> The bill is
everything around it, and it has three lines:</p>
<p><strong>server + models + your time.</strong></p>
<p>This page prices each line from the sources themselves, read on 20 October 2026: the
server from the providers&rsquo; own price pages, the models from OpenRouter&rsquo;s price list and
OpenClaw&rsquo;s own documentation of what a turn sends, the hidden lines from the same docs.
The model line is worked out in tokens first, so the arithmetic survives the next price
change. Aplite is one of the scenarios below, and this page is written by the company
behind it.</p>
<h2 id="seven-ways-to-run-it-one-month-each">Seven ways to run it, one month each</h2>
<p>Server prices are per month as printed, before tax unless the page says otherwise. The
model column assumes the example worked out in <a href="https://aplite.ai/blog/openclaw-cost-per-month#the-model-line-what-a-turn-costs">the model line</a>:
thirty messages a day on a cheap or a frontier model. &ldquo;Your time&rdquo; is what you do yourself.</p>
<div class="tscroll"><table>
<thead>
<tr>
<th>Scenario</th>
<th>Server, a month</th>
<th>Models, a month</th>
<th>Your time</th>
<th>What ends it</th>
</tr>
</thead>
<tbody>
<tr>
<td>Your own machine + <a href="https://docs.openclaw.ai/providers/ollama">Ollama</a></td>
<td>$0 plus electricity: a <a href="https://support.apple.com/103253">Mac mini M4</a> draws 4 W idle and 65 W flat out, which is $0.50 to $9 a month at US rates, €1 to €18 in Germany</td>
<td>$0</td>
<td>install, updates, a machine that never sleeps</td>
<td>sleep, reboots and network drops stop the gateway, per the <a href="https://docs.openclaw.ai/help/faq-first-run/providers-and-hosting">upstream FAQ</a></td>
</tr>
<tr>
<td><a href="https://docs.oracle.com/en-us/iaas/Content/FreeTier/freetier_topic-Always_Free_Resources.htm">Oracle Always Free</a> + an API model</td>
<td>$0: 2 OCPUs and 12 GB of Arm memory; a card at sign-up, not charged</td>
<td>$1.50 (Haiku) to $31 (Sonnet) for thirty messages a day</td>
<td>VM, firewall, install, updates</td>
<td>Oracle reclaims an instance idle for 7 days under 20 % load; &ldquo;out of host capacity&rdquo; at creation</td>
</tr>
<tr>
<td>A VPS + an API model</td>
<td>$6 to $12: <a href="https://www.hostinger.com/vps-hosting">Hostinger KVM 1</a> $6.49 paid upfront, renews at $11.99; <a href="https://www.digitalocean.com/pricing/droplets">DigitalOcean</a> $6 for 1 GiB, $12 for 2 GiB; <a href="https://contabo.com/en/vps/">Contabo</a> €5.50 incl. VAT on a 24-month term</td>
<td>the same $1.50 to $31</td>
<td>install, updates, backups</td>
<td>nothing; the renewal price</td>
</tr>
<tr>
<td>A managed host, your API key</td>
<td>$5.49 to $19.99: Flux, Railway, OneClickClaw, xCloud, MyClaw and others in <a href="https://aplite.ai/blog/managed-openclaw-hosting-compared">the comparison</a></td>
<td>the same $1.50 to $31, paid to your provider</td>
<td>agents and channels</td>
<td>—</td>
</tr>
<tr>
<td>A managed host, models included</td>
<td>$0 to $29: Runware ($0 hosting, tokens per use), Hostinger ($4.99 on 24 months, credits included), OpenClaw Launch ($6, $1 or $10 of credits), ClawHosters (€19, three free models), Everpod ($29 incl. $10), RapidClaw ($29 incl. $20)</td>
<td>included up to the allowance, then per token</td>
<td>agents and channels</td>
<td>the allowance</td>
</tr>
<tr>
<td><a href="https://aplite.ai/docs/plans">Aplite Free</a></td>
<td>$0</td>
<td>300 welcome credits, 1 credit = 1¢, budget models; no card</td>
<td>agents and channels</td>
<td>the workspace sleeps after 6 hours idle; spend is capped at 25 credits a day until you buy a pack; the 300 credits do not expire, and packs from $5 work on Free</td>
</tr>
<tr>
<td><a href="https://aplite.ai/docs/plans">Aplite Pro</a></td>
<td>$20</td>
<td>2,000 credits a month, the full catalog, credit packs from $5</td>
<td>agents and channels</td>
<td>—</td>
</tr>
</tbody>
</table></div>
<p>The rest of the page is each line in detail, with the quotes.</p>
<h2 id="the-server-line">The server line</h2>
<p><strong>Your own machine.</strong> The upstream FAQ puts the floor at &ldquo;1 vCPU, 1 GB RAM, ~500 MB disk&rdquo;
and recommends 2 GB or more; a laptop&rsquo;s advantage is listed as &ldquo;no server cost&rdquo;, its
disadvantages as &ldquo;sleep/network drops disconnect it, OS updates/reboots interrupt it,
must stay awake&rdquo;. The electricity is real but small on a small machine: Apple rates the
<a href="https://support.apple.com/103253">Mac mini M4</a> at 4 W idle and 65 W maximum, which over a
720-hour month is 2.9 to 47 kWh. At the US residential average of
<a href="https://www.eia.gov/electricity/monthly/epm_table_grapher.php?t=epmt_5_6_a">18.31 ¢/kWh</a>
(July 2026) that is $0.53 to $8.57; at Germany&rsquo;s
<a href="https://ec.europa.eu/eurostat/statistics-explained/index.php?title=Electricity_price_statistics">€0.3869</a>
(second half of 2025) it is €1.11 to €18.11, and at the EU average of €0.2896, €0.83 to
€13.55. A desktop with a GPU is a different bill: NVIDIA rates the
<a href="https://www.nvidia.com/en-us/geforce/graphics-cards/50-series/rtx-5090/">RTX 5090</a> at
575 W of graphics power alone, which would be 414 kWh, about $76 or €160, if it ran flat
out all month; it will not, but the idle draw of the whole box is what you pay for 24/7.</p>
<p>Running the model locally is what makes this scenario $0, and it is where the upstream
FAQ is blunt: asked whether local models are a good fit, it answers &ldquo;Usually no. OpenClaw
needs large context + strong safety; small cards truncate context and skip provider-side
safety filters&rdquo;, adds that &ldquo;smaller/quantized models raise prompt-injection risk&rdquo;, and
says to run &ldquo;the largest model build you can locally&rdquo;. Size is memory: Ollama&rsquo;s
<a href="https://ollama.com/library/gemma4">Gemma 4</a> weighs 6.6 to 9.5 GB at the E4B size, 16 to
19 GB at 26B and 19 to 20 GB at 31B, before the context window.</p>
<p><strong>Oracle Always Free.</strong> The allowance is &ldquo;2 OCPUs and 12 GB of memory&rdquo; on the Arm shape,
or two AMD micro instances with 1 GB each, which is exactly the floor. The clause that
ends it: &ldquo;Idle Always Free compute instances may be reclaimed by Oracle&rdquo; when, over seven
days, CPU, network and memory all stay under 20 %. An agent that waits for messages is
idle by that definition. The other free clouds and their expiry dates are in
<a href="https://aplite.ai/blog/openclaw-hosting-for-free">OpenClaw hosting for free</a>.</p>
<p><strong>A VPS.</strong> The three the assistants name most, as printed on 20 October 2026: Hostinger
KVM 1 (1 vCPU, 4 GB) at $6.49 a month, &ldquo;paid upfront&rdquo;, &ldquo;Renews at $11.99/mo for 2
years&rdquo;; DigitalOcean&rsquo;s Basic Droplet at $6.00 for 1 GiB and $12.00 for 2 GiB; Contabo
Cloud VPS 4 (4 vCPU, 8 GB) at €5.50 including VAT &ldquo;for the first 24 months&rdquo;. Hetzner,
the fourth name, renders its prices only in the browser; its cost-optimized page listed
the entry type CX23 at 2 vCPU and 4 GB and marked every type &ldquo;currently unavailable&rdquo; on
the day we read it, and its <a href="https://docs.hetzner.com/cloud/servers/overview">docs</a> add
€0.50 a month for a public IPv4 address, which the agent needs to be reached. Backups,
snapshots and a domain are the small lines on top.</p>
<p><strong>A managed host.</strong> Someone else runs the server and the install. The seventeen services
that sell this, with price from, free plan or trial, whether the key is yours and where
the servers are, fill <a href="https://aplite.ai/blog/managed-openclaw-hosting-compared">Managed OpenClaw hosting compared</a>;
the cost question splits them in two. Hosts where the model account is yours, so the
hosting price is the whole of their bill: Flux from $5.49, Railway from $5, OneClickClaw
€14.99, OpenClaw Cloud $19, BetterClaw $19, xCloud $19.99 after the first month, MyClaw
$29, ClawTrust $29. Hosts that include some model usage: Runware at $0 for hosting with
tokens billed per use, Hostinger&rsquo;s managed plan at $4.99 on a 24-month term with &ldquo;AI
credits included&rdquo; in an amount the page does not state, OpenClaw Launch at $6 with $1 or
$10 of credits, ClawHosters at €19 with three free models, Everpod at $29 including $10,
RapidClaw at $29 including $20, Agent37 Plus at $9.99, Ampere by credits. Aplite is in the
second group and has its own section below.</p>
<h2 id="the-model-line-what-a-turn-costs">The model line: what a turn costs</h2>
<p>Every message to an OpenClaw agent is at least one model call, and the call is bigger
than the message. OpenClaw&rsquo;s <a href="https://docs.openclaw.ai/concepts/context">context documentation</a>
lists what the model receives on each turn: a system prompt that is &ldquo;rendered each run&rdquo;
with the tool list, the skills list and the workspace bootstrap files (<code>AGENTS.md</code>,
<code>SOUL.md</code>, <code>IDENTITY.md</code>, <code>USER.md</code>, up to 20,000 characters each and 60,000 in total by
default), the tool schemas as JSON, which &ldquo;count toward context even though you don&rsquo;t see
them as plain text&rdquo;, the conversation history, every tool call and result, and
&ldquo;provider wrappers or hidden headers (not visible, still counted)&rdquo;. The example on that
page puts the system prompt at &ldquo;38,412 chars (~9,603 tok)&rdquo; and the tool schemas at
&ldquo;31,988 chars (~7,997 tok)&rdquo; before a single word of conversation.</p>
<p>Our own fleet agrees. Across Aplite workspaces over the sixty days to 20 October 2026,
on OpenClaw 2026.9.8 with a default workspace, the median chat turn sent about 16,000
input tokens and received about 250; the mean was 16,700 and 500. Scheduled runs were
about 14,000 in and 210 out. Roughly half of the input tokens were cache reads, which the
providers bill at a tenth of the input price or less. So the round figure for the
arithmetic below is <strong>15,000 input tokens and 400 output tokens per message</strong>, with
caching ignored, which makes every number an upper bound.</p>
<p>Three models from Aplite&rsquo;s catalog, priced from OpenRouter&rsquo;s
<a href="https://openrouter.ai/api/v1/models">model list</a> on 20 October 2026, per million tokens.
<a href="https://openrouter.ai/anthropic/claude-haiku-5.5">Claude Haiku 5.5</a> is the cheap tier.
<a href="https://openrouter.ai/deepseek/deepseek-v4-flash-0731">DeepSeek V4 Flash 0731</a> is the
platform default; OpenRouter routes it to 27 endpoints, most at $0.13 in and $0.26 to
$0.40 out, the headline figure at $0.44 and $1.32, so it is shown as a range.
<a href="https://openrouter.ai/anthropic/claude-sonnet-5.5">Claude Sonnet 5.5</a> is a frontier model.</p>
<div class="tscroll"><table>
<thead>
<tr>
<th>Model</th>
<th>Input, $/M</th>
<th>Output, $/M</th>
<th>One message (15k in, 400 out)</th>
<th>30 messages a day, 30 days</th>
</tr>
</thead>
<tbody>
<tr>
<td>Claude Haiku 5.5</td>
<td>$0.10</td>
<td>$0.50</td>
<td>0.17¢</td>
<td>$1.53</td>
</tr>
<tr>
<td>DeepSeek V4 Flash 0731</td>
<td>$0.13 to $0.44</td>
<td>$0.40 to $1.32</td>
<td>0.21¢ to 0.71¢</td>
<td>$1.90 to $6.42</td>
</tr>
<tr>
<td>Claude Sonnet 5.5</td>
<td>$2.00</td>
<td>$10.00</td>
<td>3.4¢</td>
<td>$30.60</td>
</tr>
</tbody>
</table></div>
<p>Claude Opus 5.5 at $4 and $20 doubles the Sonnet line: 6.8¢ a message, $61 a month. The
<code>:free</code> variants on OpenRouter cost $0 and are rate-limited per account to
<a href="https://openrouter.ai/docs/api-reference/limits">20 requests a minute and 50 a day</a>, or
1,000 a day once you have bought $10 of credits; one agent turn with tools is several
requests, so 50 a day is a short morning. Kilo&rsquo;s article reports one operator&rsquo;s
three-week test of the same workload on different models: $8.70 a day on Opus 4.7, $2.80
on Sonnet 4.6, $0.12 on MiniMax M2.5. The spread is the model, not the server.</p>
<p>Two things change the per-message figure and both are in your hands. Longer sessions
cost more, because the history rides along until compaction; and workspace files cost on
every turn, which is why the docs cap them. The Hostinger guide&rsquo;s observation that model
choice matters more than server size is the right conclusion from the table above.</p>
<h2 id="the-hidden-lines">The hidden lines</h2>
<p><strong>The heartbeat.</strong> By default OpenClaw wakes the agent every 30 minutes to check whether
anything needs attention, 48 times a day, and the
<a href="https://docs.openclaw.ai/gateway/heartbeat">heartbeat documentation</a> is explicit:
&ldquo;Heartbeats run full agent turns. Shorter intervals burn more tokens.&rdquo; It also gives the
size: a run in the main session carries the conversation history, about 100K tokens,
while <code>isolatedSession: true</code> and <code>lightContext: true</code> bring it down to &ldquo;~2-5K per run&rdquo;.
On our fleet the isolated, light heartbeat has a median of about 2,200 input tokens.
Output is a couple of hundred tokens and is ignored below.</p>
<div class="tscroll"><table>
<thead>
<tr>
<th>Heartbeat setting</th>
<th>Input tokens a day</th>
<th>Haiku 5.5</th>
<th>DeepSeek V4 Flash</th>
<th>Sonnet 5.5</th>
</tr>
</thead>
<tbody>
<tr>
<td>default: every 30 min in the main session, ~100K a run</td>
<td>4.8M</td>
<td>$0.48 a day, $14 a month</td>
<td>$0.62 to $2.11 a day, $19 to $63 a month</td>
<td>$9.60 a day, $288 a month</td>
</tr>
<tr>
<td>isolated and light, ~3K a run</td>
<td>144K</td>
<td>$0.01 a day, $0.43 a month</td>
<td>$0.02 to $0.06 a day, $0.56 to $1.90 a month</td>
<td>$0.29 a day, $8.64 a month</td>
</tr>
<tr>
<td>off (<code>every: "0m"</code>)</td>
<td>0</td>
<td>$0</td>
<td>$0</td>
<td>$0</td>
</tr>
</tbody>
</table></div>
<p>On a frontier model the default heartbeat costs nearly ten times the thirty messages a
day you actually send. This is the line behind the stories of an idle agent costing money: Kilo
quotes a user billed &ldquo;$5 a day for doing absolutely nothing&rdquo; on an idle GPT-5.4 agent.
The docs&rsquo; remedies are the config keys above, a cheaper <code>model</code> override for heartbeat
runs (<code>ollama/llama3.2:1b</code> is their example), a longer interval, or <code>0m</code>. Aplite&rsquo;s
runtime runs the main agent&rsquo;s heartbeat every 4 hours, isolated and light, for every
workspace, which is why the fleet figure is 2,200 and not 100,000 and six runs a day
rather than 48.</p>
<p><strong>Scheduled runs and sub-agents.</strong> A cron job is a full turn too, at the scheduled-run
size above; a daily digest is one, an hourly watcher is 24. The FAQ on
<a href="https://docs.openclaw.ai/help/faq-models">models</a> notes that sub-agents parallelize work
and &ldquo;each sub-agent consumes its own tokens&rdquo;. The Hostinger guide&rsquo;s estimate that idle or
forgotten automations are 10 to 30 % of spend is plausible against these sizes.</p>
<p><strong>Workspace files and memory.</strong> The bootstrap files are injected on every turn up to the
60,000-character cap, about 15,000 tokens if you fill it, on top of the system prompt.
A memory file that grows all month grows the bill all month.</p>
<p><strong>Seeing the line.</strong> Self-hosted, <code>/usage full</code> adds an estimated cost to each reply and
<code>/usage cost</code> prints today and the last 30 days; the
<a href="https://docs.openclaw.ai/concepts/usage-tracking">usage tracking</a> page notes that
&ldquo;sessions billed through a plan hide per-token dollar estimates&rdquo;, so a subscription login
shows tokens, not money. On Aplite the token usage shows under each reply and the money
in Settings → Cost.</p>
<p><strong>Your time.</strong> The line no provider prints. OpenClaw&rsquo;s
<a href="https://github.com/openclaw/openclaw/releases">releases</a> feed shows three stable
versions in five days, 2026.9.33 on 15 October, 2026.10.4 on 16 October and 2026.10.5 on
19 October 2026; self-hosting means deciding about each, plus the first install, the
systemd unit, the backups and the day the gateway stops answering. How much that is
worth an hour is the one number on this page only you can fill in.</p>
<h2 id="where-aplite-sits">Where Aplite sits</h2>
<p>Aplite runs a standard OpenClaw instance per workspace in Nuremberg and bills the models
in credits at 1 credit = 1¢, spent only when an agent calls a model; the credit price of a
model is above OpenRouter&rsquo;s list price, which is how the hosting is paid for on Free. The
<a href="https://aplite.ai/docs/plans">Free plan</a> is $0 with no card: one agent, 300 welcome credits that do not
expire, and the budget end of the catalog, which on 4 November means Claude Haiku 5.5,
Gemini 3.5 Flash Lite, DeepSeek V4 Flash, GLM 5.3 Flash and Qwen3.8 Flash. The workspace
sleeps after 6 hours idle and wakes on the next visit, Free has no schedules, and spend is
capped at 25 credits a day until a pack is bought; when the 300 credits are gone, a credit
pack keeps it going and Pro adds the rest. <a href="https://aplite.ai/pricing">Pro</a> is $20 a month
with 2,000 credits that reset monthly, six agents, always-on, schedules and the full
catalog; packs run from $5 for 500 credits and never expire. Power is $50 with 6,000.</p>
<p>What that buys in messages, from the fleet figures above: a chat turn on the default
model has cost 0.4 to 0.5 credits on average over the last sixty days, so the welcome
grant is roughly 600 to 750 turns and Pro&rsquo;s monthly credits roughly 4,000 to 5,000, before
any scheduled runs; a Sonnet pin multiplies the per-turn cost by five to fifteen. What
real workspaces spent, over the same sixty days: the median active day used about 5
credits and the busiest tenth of days about 22. Against the table, Pro at $20 is more
than a $6 VPS with a cheap model and less than the same VPS with a frontier model and a
default heartbeat; Free is $0 for as long as the credits last, with a gateway that is
asleep when nobody is talking to it. Aplite is run by PIKsolution OÜ, an Estonian
company, and is not affiliated with the OpenClaw project.</p>
<p>Prices on this page move. Every line carries the date it was read; the plan is to re-read
them once a quarter.</p>
<h2 id="faq">FAQ</h2>
<h3 id="is-openclaw-free">Is OpenClaw free?</h3>
<p>The software is: OpenClaw is open source under the MIT license, and its site says &ldquo;No
subscription. No hosted tier. No token.&rdquo; What costs money is the machine it runs on and
the model it calls. The only setup with both at $0 is your own computer running a local
model through Ollama, which the upstream FAQ warns is &ldquo;usually no&rdquo; for a good fit because
small models truncate context; every other path pays for the server, the tokens or both.</p>
<h3 id="how-much-do-the-models-cost-per-month">How much do the models cost per month?</h3>
<p>It depends on the model and on the heartbeat more than on how much you type. At thirty
messages a day, with each message sending about 15,000 tokens of context, a month on
Claude Haiku 5.5 is about $1.50, on DeepSeek V4 Flash $2 to $6.50, on Claude Sonnet 5.5
about $31, on Claude Opus 5.5 about $61, at OpenRouter&rsquo;s prices on 20 October 2026. The
default heartbeat, every 30 minutes in the main session at about 100K tokens a run, adds
$14 a month on Haiku and $288 on Sonnet unless it is isolated or switched off.</p>
<h3 id="what-is-the-cheapest-way-to-run-openclaw-247">What is the cheapest way to run OpenClaw 24/7?</h3>
<p>In money, your own always-on machine with a local model: $0 plus electricity, which is
under $1 a month for a Mac mini at idle. If the machine must be someone else&rsquo;s, Oracle&rsquo;s
Always Free tier is $0 for a 2-OCPU, 12 GB Arm instance with a card at sign-up, and
reclaims instances idle for seven days; the cheapest paid servers are $5 to $6.50 a month
at Contabo, DigitalOcean and Hostinger, with renewals up to $11.99. The model tokens come
on top in every case, from about $1.50 a month on a cheap model. Managed hosts with a
free plan, Aplite among them, are $0 for the hosting with a credit allowance for the
models; which of them keep a free plan and which take a card is in
<a href="https://aplite.ai/blog/openclaw-free-plan-no-credit-card">OpenClaw hosting with a free plan and no credit card</a>.</p>
<h3 id="how-many-tokens-does-one-openclaw-message-use">How many tokens does one OpenClaw message use?</h3>
<p>Far more than the message. OpenClaw&rsquo;s context documentation shows a system prompt of
about 9,600 tokens and tool schemas of about 8,000 before any conversation, and on
Aplite&rsquo;s fleet the median chat turn in October 2026 sent about 16,000 input tokens and
received about 250. Half of the input is typically served from cache at a fraction of the
price. A good planning figure is 15,000 in and 400 out per message.</p>
<h3 id="why-is-my-openclaw-bill-so-high">Why is my OpenClaw bill so high?</h3>
<p>Usually the heartbeat: by default the agent runs a full turn every 30 minutes, 48 a day,
and in the main session each one carries the history, about 100K tokens. On a frontier
model that is several dollars a day with nobody talking to the agent. Set
<code>isolatedSession: true</code> and <code>lightContext: true</code>, give heartbeat runs a cheap model, or
disable them with <code>every: "0m"</code>. After that, look at scheduled jobs and sub-agents, which
are full turns too, and at the size of the workspace files injected on every turn.</p>
<h3 id="how-much-does-aplite-cost-per-month">How much does Aplite cost per month?</h3>
<p>Free is $0 with no card: one agent, 300 welcome credits that do not expire, budget
models, 25 credits a day until a pack is bought, and a workspace that sleeps after 6 hours
idle. Pro is $20 a month with 2,000
credits, Power $50 with 6,000; 1 credit is 1¢ of model usage, priced above OpenRouter&rsquo;s
list. On the default model a chat turn has averaged 0.4 to 0.5 credits across our
workspaces, so Pro&rsquo;s credits are several thousand turns a month before scheduled runs.</p>]]></content:encoded>
  </item>
  <item>
    <title>OpenClaw install on Windows and Mac: what the installer does, where it breaks, and the hosted way around it</title>
    <link>https://aplite.ai/blog/openclaw-install-windows-mac</link>
    <guid isPermaLink="true">https://aplite.ai/blog/openclaw-install-windows-mac</guid>
    <pubDate>Tue, 20 Oct 2026 00:00:00 +0000</pubDate>
    <description>What install.sh and install.ps1 actually do, the Node and OS requirements as of October 2026, seven real error messages with their fixes, a post-install checklist, and the way to skip the install.</description>
    <content:encoded><![CDATA[<figure class="fig">
<a class="zoom" href="https://aplite.ai/assets/blog/openclaw-install-windows-mac.webp?v=00b05817" target="_blank" rel="noopener"><img src="https://aplite.ai/assets/blog/openclaw-install-windows-mac.webp?v=00b05817" alt="Two installers, one gateway: the Mac path through Homebrew and a LaunchAgent, the Windows path through winget and a Scheduled Task." width="1560" height="820" decoding="async"></a>
<figcaption>Two installers, one gateway: the Mac path through Homebrew and a LaunchAgent, the Windows path through winget and a Scheduled Task.</figcaption>
</figure>
<p>Search for <em>openclaw install windows</em> or <em>openclaw install mac</em> and the first page is
vendors: Kimi&rsquo;s guide ends at its own cloud and still says Node 22.14, Parallels wants
you in a VM, hosting blogs want you on their server. The official answer is spread over
five docs pages. This one puts it in order: what the two installer scripts do step by
step, the requirements that changed in September, the errors people hit with their
real text, and a checklist for the end. Everything about OpenClaw is from the
<a href="https://docs.openclaw.ai/install">OpenClaw docs</a> and the scripts themselves, read on
20 October 2026 (release 2026.10.5).</p>
<h2 id="requirements-as-of-october-2026">Requirements, as of October 2026</h2>
<ul>
<li><strong>Node 24.16 or newer, or 26.1 or newer.</strong> Node 22, 23 and 25 are unsupported since
  2026.9.3, because their <code>node:sqlite</code> silently truncates text at an embedded NUL
  (<a href="https://docs.openclaw.ai/install/node-compatibility">Node.js compatibility</a>).
  The installer adds Node when it is missing: 26 via Homebrew on a Mac, the LTS line
  via winget on Windows.</li>
<li><strong>Windows:</strong> PowerShell 5 or newer for the script; Windows 10 20H2 or 11 for the
  Windows Hub app. WSL2 is optional, not required
  (<a href="https://docs.openclaw.ai/platforms/windows">Windows</a>).</li>
<li><strong>Mac:</strong> the CLI runs where Node runs, macOS 13.5 or newer; the menu bar app needs
  macOS 26.2 Tahoe (<a href="https://docs.openclaw.ai/platforms/macos">macOS app</a>).</li>
<li><strong>Git</strong>, even for an npm install: the installer checks for it so dependencies from
  git URLs do not fail with <code>spawn git ENOENT</code>.</li>
<li><strong>A model:</strong> a provider API key, a Claude Code or Codex login, or Ollama. Every
  route, including the free ones: <a href="https://aplite.ai/blog/openclaw-without-api-key">OpenClaw without an API key</a>.</li>
</ul>
<h2 id="which-path">Which path</h2>
<p>On Windows the docs recommend the <strong>Windows Hub</strong> app
(<a href="https://github.com/openclaw/openclaw-windows-node/releases/latest">OpenClawCompanion-Setup-x64.exe</a>,
2026.9.8-1 on 7 October): no administrator rights, and <em>Set up locally</em> creates its own
WSL distro, <code>OpenClawGateway</code>. The <strong>PowerShell script</strong> gives a native CLI and a
gateway kept alive by a Scheduled Task; <strong>WSL2</strong> runs the Mac/Linux script. On a Mac:
the <strong>menu bar app</strong> (2026.10.5 ships Apple Silicon, Intel and Universal DMGs) or the
<strong>curl script</strong>, which installs a LaunchAgent. This page follows the scripts.</p>
<h2 id="what-installsh-does-on-a-mac">What install.sh does on a Mac</h2>
<p>About five minutes by the <a href="https://docs.openclaw.ai/start/getting-started">getting started</a>
page. The flow, from <a href="https://docs.openclaw.ai/install/installer">installer internals</a>
and the script:</p>
<ol>
<li>
<p><strong>Run it</strong> in Terminal:</p>
<pre><code>curl -fsSL https://openclaw.ai/install.sh | bash
</code></pre>
</li>
<li>
<p><strong>Node.</strong> The script checks the version and the linked SQLite. If Node is missing it
   installs Homebrew first (that needs an Administrator account) and then <code>brew install
   node</code>, which is Node 26. An nvm install is reused when compatible; with nvm present
   but no compatible version it offers <code>nvm install 26</code> and never installs a second nvm.</p>
</li>
<li>
<p><strong>Git</strong> via Homebrew if missing, then <code>npm install -g openclaw@latest</code>, with
   <code>--allow-scripts=openclaw</code> added on npm 11.16 and newer. A failed build pulls in
   Xcode Command Line Tools and retries.</p>
</li>
<li>
<p><strong>Onboarding.</strong> On a fresh install it hands over to <code>openclaw onboard</code>; <em>Quick
   start</em> finds a Claude Code or Codex login or a key, checks it with a real
   completion and opens the dashboard. On an existing install it runs
   <code>openclaw doctor --fix</code> and restarts the gateway service instead.</p>
</li>
<li>
<p><strong>Keep it running.</strong> Quick start leaves the gateway in the foreground. Press
   Ctrl+C, then:</p>
<pre><code>openclaw gateway install
openclaw gateway status
</code></pre>
<p>The first installs a LaunchAgent; the second should show the gateway listening on
port 18789.</p>
</li>
</ol>
<h2 id="what-installps1-does-on-windows">What install.ps1 does on Windows</h2>
<ol>
<li>
<p><strong>Run it</strong> in PowerShell (no execution-policy change needed for this form):</p>
<pre><code>iwr -useb https://openclaw.ai/install.ps1 | iex
</code></pre>
</li>
<li>
<p><strong>Node.</strong> If missing: winget, then Chocolatey, then Scoop. If all three are absent
   or leave an unsupported version, the script downloads the official Node 26 zip into
   <code>%LOCALAPPDATA%\OpenClaw\deps\portable-node</code> and adds it to your user PATH.</p>
</li>
<li>
<p><strong>OpenClaw</strong> by <code>npm install -g</code>, launched from a temp directory so a shell opened in
   <code>C:\</code> still works. If Git is missing on the git install method, a user-local MinGit
   goes under <code>%LOCALAPPDATA%\OpenClaw\deps\portable-git</code>.</p>
</li>
<li>
<p><strong>PATH and doctor.</strong> The bin directory goes on your user PATH; on upgrades it runs
   <code>openclaw doctor --fix --non-interactive</code>. A failure stops with a terminating error
   and leaves the window open.</p>
</li>
<li>
<p><strong>Onboarding</strong>, as on the Mac. Then the service:</p>
<pre><code>openclaw gateway install
openclaw gateway status --json
</code></pre>
<p>This registers a Scheduled Task, &ldquo;OpenClaw Gateway&rdquo;, that runs as you before login
without storing a password. If task creation is denied you get a Startup-folder
item instead, which only starts after you log in; run the install from an elevated
terminal to get the boot task.</p>
</li>
</ol>
<h2 id="where-it-breaks">Where it breaks</h2>
<p>The error text as it appears, why, and the fix, with the source.</p>
<ul>
<li><strong><code>openclaw is not recognized</code></strong> on Windows, <strong><code>zsh: command not found: openclaw</code></strong> on
  a Mac. npm&rsquo;s global bin directory is not on PATH, or the terminal predates the
  install. Reopen the terminal; still missing: <code>npm config get prefix</code> and add that
  directory to the user PATH on Windows (no <code>\bin</code>), on a Mac
  <code>export PATH="$(npm prefix -g)/bin:$PATH"</code> in <code>~/.zshrc</code>
  (<a href="https://docs.openclaw.ai/install/node#troubleshooting">docs</a>).</li>
<li><strong><code>npm error spawn git ENOENT</code></strong> on Windows. Git is missing, so a dependency from a
  git URL cannot be cloned. Rerun the script so it bootstraps MinGit, or install Git for
  Windows and reopen PowerShell
  (<a href="https://docs.openclaw.ai/install/installer#troubleshooting">docs</a>). January&rsquo;s
  <code>node.exe : npm error code ENOENT</code> was the installer calling npm through <code>node.exe</code>
  (<a href="https://github.com/openclaw/openclaw/issues/3917">#3917</a>,
  <a href="https://github.com/openclaw/openclaw/issues/11049">#11049</a>); gone since.</li>
<li><strong><code>Error: Could not find a package manager (winget, choco, or scoop)</code></strong>. Windows
  without winget, and the window closed before you could read it
  (<a href="https://github.com/openclaw/openclaw/issues/38054">#38054</a>, March). Fixed since:
  the script downloads portable Node, and the piped form keeps the window open.
  If it recurs, install winget from the Store or Node from nodejs.org.</li>
<li><strong><code>Node 22.22.0: node:sqlite truncates TEXT at embedded NUL (nodejs/node#61954); use
  24.16+/26.1+ or a build with the fix</code></strong>, with a <code>v</code> before the number on Windows.
  Node below the September floor. Let the installer provision Node, or answer <strong>Y</strong>
  to the CLI&rsquo;s <code>Update NodeJS: Y/N [N]:</code> prompt; a private runtime lands in
  <code>~/.openclaw/tools/cli-node</code>.</li>
<li><strong><code>blocked because they are not covered by allowScripts</code></strong>. A manual
  <code>npm install -g openclaw</code> on npm 12. Add <code>--allow-scripts=openclaw</code>;
  <code>npm approve-scripts</code> does not work for global installs
  (<a href="https://docs.openclaw.ai/install#npm-pnpm-or-bun">docs</a>).</li>
<li><strong><code>npm.ps1 cannot be loaded because running scripts is disabled on this system</code></strong>,
  same for <code>openclaw.ps1</code>. Running npm or openclaw by hand under the Restricted policy
  Windows client editions ship with; npm&rsquo;s <code>.ps1</code> shims are scripts, the piped
  installer is not. <code>Set-ExecutionPolicy RemoteSigned -Scope CurrentUser</code>, or call
  <code>npm.cmd</code> and <code>openclaw.cmd</code>
  (<a href="https://learn.microsoft.com/en-us/powershell/module/microsoft.powershell.core/about/about_execution_policies">Microsoft</a>).</li>
<li><strong><code>You have not agreed to the Xcode license agreements</code></strong>, or the &ldquo;install the
  command line developer tools&rdquo; dialog on a Mac. npm needs git; without Command Line
  Tools, Apple&rsquo;s <code>/usr/bin/git</code> shim opens the dialog, and the gateway runs git at
  every start even with update checks off
  (<a href="https://github.com/openclaw/openclaw/issues/31527">#31527</a>,
  <a href="https://github.com/openclaw/openclaw/issues/158099">#158099</a>, open).
  <code>xcode-select --install</code>, finish the dialog, rerun; <code>sudo xcodebuild -license</code> only
  if full Xcode is installed.</li>
</ul>
<p>Two more, after the install rather than during it. On Windows a Scheduled Task that
reports success proves nothing, &ldquo;A successful task result alone does not prove the
Gateway is healthy&rdquo;: read <code>openclaw gateway status --json</code> and the gateway log, as the
2026.9.1 update taught (<a href="https://github.com/openclaw/openclaw/issues/137813">#137813</a>).
On a Mac an open issue
has the LaunchAgent reach <code>ready</code>, get SIGTERM a second later and loop, while the same
build is stable in the foreground
(<a href="https://github.com/openclaw/openclaw/issues/144094">#144094</a>, 2026.8.2 and 9.3);
<code>openclaw gateway run</code> tells the two apart. On 2026.9.4 on Windows, read the docs&rsquo;
<a href="https://docs.openclaw.ai/platforms/windows#updating-from-2026-9-4">note on updating</a>
before <code>openclaw update</code>.</p>
<h2 id="after-the-install-a-checklist">After the install: a checklist</h2>
<ol>
<li><code>openclaw --version</code> prints 2026.10.10 or whatever you asked for.</li>
<li><code>openclaw doctor</code> has nothing red; <code>openclaw doctor --fix</code> repairs what it can.</li>
<li><code>openclaw gateway status</code> shows the gateway on port 18789, from a new terminal.</li>
<li><code>openclaw dashboard</code> opens the Control UI and a typed message gets a reply.</li>
<li>The gateway is bound to <code>127.0.0.1</code>, which is the default: do not forward 18789
   on your router (<a href="https://docs.openclaw.ai/gateway/security">security</a>).</li>
<li>A first channel: <a href="https://aplite.ai/blog/openclaw-telegram-setup">OpenClaw in Telegram in five minutes</a>.
   An unknown sender gets a pairing code until you approve them.</li>
<li>The machine stays on. Sleep, a closed lid and OS updates stop the gateway; the
   upstream FAQ says to use a VPS after disconnects
   (<a href="https://docs.openclaw.ai/help/faq-first-run/providers-and-hosting">providers and hosting</a>).</li>
<li>When something is off, <code>openclaw triage</code> runs read-only checks and writes a
   diagnosis for Claude Code, Codex or the built-in agent.</li>
</ol>
<h2 id="without-installing-anything">Without installing anything</h2>
<p><a href="https://aplite.ai/docs/openclaw">Aplite</a> runs a standard OpenClaw instance for you: one gateway in its own
container on our server in Nuremberg, currently 2026.10.10, set up from the browser.</p>
<ol>
<li>Create an account with email or Google; no card on the Free plan.</li>
<li>Pick an agent and press <strong>Deploy</strong>; the workspace opens under a minute later.</li>
<li>Ask — in the browser, or from your phone through Telegram, WhatsApp or WeChat. There
   is no Node, no PATH, no key to paste: models come through one metered
   account, billed in credits, 1 credit = 1¢, token usage under each reply.</li>
</ol>
<p>What a managed instance gives up, the shell, raw <code>openclaw.json</code>, local models and
day-one releases, is in <a href="https://aplite.ai/blog/managed-openclaw-vs-self-hosting">Managed OpenClaw vs self-hosting</a>;
other hosts are in <a href="https://aplite.ai/blog/managed-openclaw-hosting-compared">Managed OpenClaw hosting compared</a>.
Aplite is independent and not affiliated with the OpenClaw project.</p>
<h2 id="faq">FAQ</h2>
<h3 id="does-openclaw-run-on-windows">Does OpenClaw run on Windows?</h3>
<p>Yes, three ways: the Windows Hub desktop app, which creates its own WSL distro; the
PowerShell installer, a native CLI and a gateway kept alive by a Scheduled Task; or
WSL2 with the Linux script. The docs recommend the Hub for desktop use.</p>
<h3 id="do-i-need-wsl2-to-install-openclaw-on-windows">Do I need WSL2 to install OpenClaw on Windows?</h3>
<p>No. The PowerShell script installs a native gateway. WSL2 is the most Linux-compatible
option and what the Hub uses underneath, but it is not required.</p>
<h3 id="which-node-version-does-openclaw-need">Which Node version does OpenClaw need?</h3>
<p>Node 24.16 or newer, or 26.1 or newer; Node 26 is recommended. Node 22, 23 and 25 are
unsupported since release 2026.9.3. The installer provisions a supported Node when
yours is missing or too old.</p>
<h3 id="does-openclaw-work-on-an-intel-mac-or-an-older-macos">Does OpenClaw work on an Intel Mac or an older macOS?</h3>
<p>The CLI and gateway run wherever Node 24 or 26 runs, which Node builds for macOS 13.5
and newer, on Intel and Apple Silicon. The menu bar app needs macOS 26.2 Tahoe; on an
older Mac use the script.</p>
<h3 id="why-does-powershell-say-openclaw-is-not-recognized-after-the-install">Why does PowerShell say openclaw is not recognized after the install?</h3>
<p>The global npm bin directory is not on the PATH of that window. Open a new PowerShell
first; if it still fails, run <code>npm config get prefix</code> and add that directory to your
user PATH, then reopen.</p>
<h3 id="can-i-try-openclaw-without-installing-it">Can I try OpenClaw without installing it?</h3>
<p>Yes. A managed host runs the gateway for you; on Aplite that is an account, Deploy and
a chat, on the Free plan without a card. The trade-offs are above.</p>]]></content:encoded>
  </item>
  <item>
    <title>OpenClaw in Telegram in five minutes: setup, pairing, hosted or self-hosted</title>
    <link>https://aplite.ai/blog/openclaw-telegram-setup</link>
    <guid isPermaLink="true">https://aplite.ai/blog/openclaw-telegram-setup</guid>
    <pubDate>Tue, 20 Oct 2026 00:00:00 +0000</pubDate>
    <description>A BotFather token, one command, one approve. OpenClaw&#x27;s Telegram channel on 2026.9.8 step by step, and the same bot on a hosted workspace with no server.</description>
    <content:encoded><![CDATA[<figure class="fig">
<a class="zoom" href="https://aplite.ai/assets/blog/openclaw-telegram-setup.webp?v=9c088a7b" target="_blank" rel="noopener"><img src="https://aplite.ai/assets/blog/openclaw-telegram-setup.webp?v=9c088a7b" alt="Self-hosted: one command adds the token, one approves the chat. Hosted: the same approve is a button." width="1560" height="820" decoding="async"></a>
<figcaption>Self-hosted: one command adds the token, one approves the chat. Hosted: the same approve is a button.</figcaption>
</figure>
<p>Search for <em>openclaw telegram setup</em> and you get a dozen walkthroughs by hosting companies,
each ending at their own control panel, and the upstream documentation, which is correct but
spread over four pages: setup, pairing, access control, troubleshooting. The actual job is
short. A Telegram bot is a token from @BotFather; OpenClaw needs that token and then an
approval for every chat that may talk to it. That is the whole thing, and it takes about
five minutes whether the gateway runs on your machine or on someone else&rsquo;s.</p>
<p>Below, both paths on OpenClaw 2026.9.8: the self-hosted one with every command, checked
against the <a href="https://docs.openclaw.ai/channels/telegram/setup">upstream docs</a> and run in a
fresh container of that version, then the <a href="https://aplite.ai/blog/openclaw-telegram-setup#hosted-the-same-bot-in-three-steps">hosted one</a>,
where the commands become two buttons.</p>
<h2 id="what-you-need">What you need</h2>
<ul>
<li><strong>A bot token.</strong> Free, from <a href="https://t.me/BotFather">@BotFather</a> inside Telegram. Check
  the handle is exactly <code>@BotFather</code>; there are impostors.</li>
<li><strong>A gateway that stays up.</strong> The Telegram channel long-polls Telegram&rsquo;s API from the
  gateway, so a bot only answers while the gateway runs. A laptop that sleeps stops it.</li>
<li><strong>A model.</strong> The bot token is not a model key. Self-hosted, you still need a provider or
  a local model (<a href="https://aplite.ai/blog/openclaw-without-api-key">the ways without an API key</a>); hosted,
  the models are included.</li>
</ul>
<h2 id="self-hosted-six-steps">Self-hosted: six steps</h2>
<ol>
<li>
<p><strong>Create the bot.</strong> In Telegram, open @BotFather, send <code>/newbot</code>, answer with a display
   name and a username ending in <code>bot</code>. BotFather replies with the token, shaped like
   <code>123456789:AAF…</code>. Keep it private: whoever has it can run your bot until you regenerate
   it with <code>/revoke</code>.</p>
</li>
<li>
<p><strong>Install OpenClaw</strong> if it is not there yet. On macOS, Linux or WSL2:</p>
<pre><code>curl -fsSL https://openclaw.ai/install.sh | bash
openclaw onboard
</code></pre>
<p>Onboarding picks the model provider and starts the gateway; the longer version is in
<a href="https://aplite.ai/blog/openclaw-hosting-for-free#the-0-path-step-by-step">OpenClaw hosting for free</a>.</p>
</li>
<li>
<p><strong>Give OpenClaw the token.</strong> One command:</p>
<pre><code>openclaw channels add --channel telegram --token 123456789:AAF…
</code></pre>
<p>It prints <code>Added Telegram account "default".</code> and writes the channel into
<code>~/.openclaw/openclaw.json</code>:</p>
<pre><code>"channels": {
  "telegram": {
    "enabled": true,
    "botToken": "123456789:AAF…"
  }
}
</code></pre>
<p>The same block typed by hand works too, as does the <code>TELEGRAM_BOT_TOKEN</code> environment
variable for the default account; a <code>tokenFile</code> beats <code>botToken</code>, which beats the env.
There is no <code>openclaw channels login telegram</code> step: unlike WhatsApp with its QR
login, Telegram needs nothing but the token
(<a href="https://docs.openclaw.ai/channels/telegram/setup">setup docs</a>).</p>
</li>
<li>
<p><strong>Make sure the gateway is running.</strong> A running gateway hot-reloads the new channel. If
   it is not running, start it in the foreground with <code>openclaw gateway</code>, or as the service
   that <code>openclaw gateway install</code> set up (<a href="https://docs.openclaw.ai/gateway">gateway docs</a>):</p>
<pre><code>openclaw gateway start
</code></pre>
<p>Then check the channel:</p>
<pre><code>openclaw channels status --probe
</code></pre>
<p>The Telegram line reads <code>Telegram default: enabled, configured, mode:polling,
token:***</code>; with <code>--probe</code> the gateway also calls Telegram with the token. A wrong
token fails at gateway startup as <code>getMe returned 401</code>.</p>
</li>
<li>
<p><strong>Send <code>/start</code> to your bot.</strong> A Telegram bot cannot write to you first. On the first
   message from an unknown chat, OpenClaw&rsquo;s default DM policy, <code>pairing</code>, does not answer
   the question; it replies with a pairing code instead: eight characters, uppercase, no
   <code>0</code>, <code>O</code>, <code>1</code> or <code>I</code>, valid for one hour, at most three pending per bot
   (<a href="https://docs.openclaw.ai/channels/pairing">pairing docs</a>).</p>
</li>
<li>
<p><strong>Approve the chat.</strong> On the server:</p>
<pre><code>openclaw pairing list telegram
openclaw pairing approve telegram K7QX2MPB
</code></pre>
<p>Add <code>--notify</code> and the bot tells the requester they are in. From now on that chat talks
to the agent; the approval lives in <code>~/.openclaw/state/openclaw.sqlite</code>, and the first
approved sender becomes the command owner if none was set
(<a href="https://docs.openclaw.ai/cli/pairing">CLI reference</a>). Anyone else who finds the bot
gets a code and silence until you approve them, which is the point.</p>
</li>
</ol>
<p>Two things people do next, both optional. <strong>Lock it to yourself</strong>: set <code>dmPolicy</code> to
<code>allowlist</code> and put your numeric Telegram user ID in <code>allowFrom</code>. The ID is in the bot&rsquo;s
pairing reply, in <code>openclaw logs --follow</code> as <code>senderUserId</code>, or at
<code>https://api.telegram.org/bot&lt;token&gt;/getUpdates</code>; <code>@username</code> entries are not accepted
(<a href="https://docs.openclaw.ai/channels/telegram/access-control">access control</a>). <strong>Add it to a
group</strong>: groups default to <code>groupPolicy: "allowlist"</code>, so a group is silent until its chat ID
(a <code>-100…</code> number for supergroups) is listed under <code>channels.telegram.groups</code>; the bot then
answers only when mentioned unless <code>requireMention</code> is off, and reading unmentioned messages
needs Privacy Mode disabled in BotFather (<code>/setprivacy</code>) or admin rights, then a remove and
re-add. <code>/whoami@your_bot</code> in the group confirms both IDs.</p>
<h2 id="hosted-the-same-bot-in-three-steps">Hosted: the same bot in three steps</h2>
<p><a href="https://aplite.ai/docs/openclaw">Aplite</a> runs a standard OpenClaw instance for you, so the gateway is already
up and the config is a form. Same bot, same pairing rule, no server:</p>
<ol>
<li>
<p><strong>Create a workspace.</strong> Email or Google, no card on the Free plan. Pick an agent, press
   Deploy, and under a minute later the workspace opens with its own OpenClaw 2026.10.10 in
   its own container.</p>
</li>
<li>
<p><strong>Connect the bot.</strong> Open <strong>Skills → Channels → Telegram</strong> and press <strong>Connect</strong>. Paste
   the token from @BotFather, press <strong>Validate</strong>: Aplite calls Telegram&rsquo;s <code>getMe</code> and shows
   <code>Connected to @your_bot</code>. Press <strong>Connect</strong> again; the workspace restarts to pick up the
   bot, about a minute. Under <em>Show advanced</em> sit the same knobs the config file has: DM
   policy (pairing by default), group policy (allowlist), chat IDs, require @-mention.</p>
</li>
</ol>
<figure class="shot">
<span class="shot-bar" aria-hidden="true"><i></i><i></i><i></i></span>
<a class="zoom" href="https://aplite.ai/assets/blog/openclaw-telegram-setup-skills.webp?v=3ba86137" target="_blank" rel="noopener"><img src="https://aplite.ai/assets/blog/openclaw-telegram-setup-skills.webp?v=3ba86137" alt="Skills → Channels → Telegram: the card is one Connect button away from a bot." loading="lazy"></a>
<figcaption>Skills → Channels → Telegram: the card is one Connect button away from a bot.</figcaption>
</figure>
<figure class="shot">
<span class="shot-bar" aria-hidden="true"><i></i><i></i><i></i></span>
<a class="zoom" href="https://aplite.ai/assets/blog/openclaw-telegram-setup-connect.webp?v=c5291b48" target="_blank" rel="noopener"><img src="https://aplite.ai/assets/blog/openclaw-telegram-setup-connect.webp?v=c5291b48" alt="Paste the token, Validate, Connect: the dialog does what channels add does." loading="lazy"></a>
<figcaption>Paste the token, Validate, Connect: the dialog does what channels add does.</figcaption>
</figure>
<ol start="3">
<li><strong>Say <code>/start</code>, then Approve.</strong> Send <code>/start</code> to the bot from your phone. A pairing
   request appears on the same Telegram card within seconds, with the sender&rsquo;s name and the
   code; press <strong>Approve</strong>. Codes expire after one hour, so if you were slow, send <code>/start</code>
   again. Only chats you approve can talk to your agent.</li>
</ol>
<figure class="fig">
<a class="zoom" href="https://aplite.ai/assets/blog/openclaw-telegram-setup-policies.webp?v=b045b2e4" target="_blank" rel="noopener"><img src="https://aplite.ai/assets/blog/openclaw-telegram-setup-policies.webp?v=b045b2e4" alt="The advanced block: pairing, allowlist, open or disabled for DMs; allowlist for groups; the mention rule." width="504" height="1084" decoding="async"></a>
<figcaption>The advanced block: pairing, allowlist, open or disabled for DMs; allowlist for groups; the mention rule.</figcaption>
</figure>
<p>That is the setup. The bot answers in Telegram with the same agent that answers in the
browser, routines can <a href="https://aplite.ai/blog/ai-assistant-in-telegram">deliver to it</a>, and the toggle on the
Telegram card pauses the channel without dropping the token; WhatsApp and WeChat sit on the
same Channels screen and link by QR. One honest line about the Free
plan: the workspace sleeps after 6 hours idle and wakes on your next visit, and while it
sleeps the bot does not answer, so a bot that answers at any hour needs
<a href="https://aplite.ai/docs/plans">Pro or Power</a>. Aplite is an independent service and is not affiliated with the
OpenClaw project.</p>
<h2 id="if-the-bot-stays-silent">If the bot stays silent</h2>
<ul>
<li><strong>Nothing at all after <code>/start</code>.</strong> It is waiting for approval. Run
  <code>openclaw pairing list telegram</code>, approve the code, or approve on the Telegram card. A
  code older than an hour is gone; send <code>/start</code> again
  (<a href="https://docs.openclaw.ai/channels/troubleshooting">channel troubleshooting</a>).</li>
<li><strong><code>getMe returned 401</code> or <code>deleteWebhook 401 Unauthorized</code> at startup.</strong> The token is
  wrong or revoked. Copy it again from BotFather and update <code>botToken</code>, <code>tokenFile</code> or
  <code>TELEGRAM_BOT_TOKEN</code>
  (<a href="https://docs.openclaw.ai/channels/telegram/troubleshooting">Telegram troubleshooting</a>).</li>
<li><strong>Answers in DMs, silent in a group.</strong> The group is not in <code>channels.telegram.groups</code>, or
  Privacy Mode is on while you expect unmentioned messages: <code>/setprivacy</code> → Disable, then
  remove and re-add the bot. Hosted, the same two settings are in the advanced block.</li>
<li><strong>Worked yesterday, silent today.</strong> Self-hosted: <code>openclaw gateway status</code> should say
  <code>Runtime: running</code>; a laptop that slept or a VPS that rebooted without
  <code>openclaw gateway install</code> stops polling. Hosted on Free: the workspace is asleep; open it
  once.</li>
</ul>
<h2 id="faq">FAQ</h2>
<h3 id="does-openclaw-need-a-server-to-work-in-telegram">Does OpenClaw need a server to work in Telegram?</h3>
<p>Yes. The gateway polls Telegram for new messages, so something has to keep running: your
machine while it is awake, a VPS, or a managed instance. A message sent while the gateway
is down is not answered until it is back.</p>
<h3 id="can-more-than-one-person-use-the-bot">Can more than one person use the bot?</h3>
<p>Yes. Each person sends <code>/start</code>, gets their own pairing code, and you approve each one; an
unapproved sender never reaches the agent. For a shared group, list its chat ID under
<code>groups</code> or switch the group policy to open.</p>
<h3 id="do-i-need-an-api-key-for-the-telegram-bot">Do I need an API key for the Telegram bot?</h3>
<p>Not for Telegram: the bot token from @BotFather is free. The model behind the bot is the
usual question. Self-hosted, that is a provider key, a subscription login or a local model
(<a href="https://aplite.ai/blog/openclaw-without-api-key">every option</a>); on Aplite the models are metered and
included.</p>
<h3 id="why-does-the-bot-answer-my-question-with-a-code">Why does the bot answer my question with a code?</h3>
<p>That is pairing, OpenClaw&rsquo;s default DM policy: the first message from an unknown chat gets
a code, not an answer, until the operator approves it with
<code>openclaw pairing approve telegram &lt;CODE&gt;</code> or the Approve button. Set <code>dmPolicy</code> to <code>open</code>
only if you want anyone who finds the bot to command your agent.</p>]]></content:encoded>
  </item>
  <item>
    <title>OpenClaw hosting in the EU: GDPR, data residency and who actually offers it</title>
    <link>https://aplite.ai/blog/openclaw-hosting-eu-gdpr</link>
    <guid isPermaLink="true">https://aplite.ai/blog/openclaw-hosting-eu-gdpr</guid>
    <pubDate>Tue, 20 Oct 2026 00:00:00 +0000</pubDate>
    <description>What the GDPR asks of the company that runs your agent: a server region, a processing agreement, a sub-processor list, and an answer to where the prompts go. 17 managed OpenClaw hosts and 4 European VPS providers in one table, read on their own legal pages on 20 October 2026.</description>
    <content:encoded><![CDATA[<figure class="fig">
<a class="zoom" href="https://aplite.ai/assets/blog/openclaw-hosting-eu-gdpr.webp?v=52a8593c" target="_blank" rel="noopener"><img src="https://aplite.ai/assets/blog/openclaw-hosting-eu-gdpr.webp?v=52a8593c" alt="Seventeen managed OpenClaw hosts and four EU VPS providers: where the servers are, who the legal entity is, whether a DPA exists, where the prompts go." width="1560" height="820" decoding="async"></a>
<figcaption>Seventeen managed OpenClaw hosts and four EU VPS providers: where the servers are, who the legal entity is, whether a DPA exists, where the prompts go.</figcaption>
</figure>
<p>&ldquo;OpenClaw hosting in the EU&rdquo; is asked for one of two reasons. A reader in Europe wants the
agent&rsquo;s notes, memory and chat history on a server under EU law. Or a company has to show a
processing agreement for every service that touches personal data, and an agent that reads a
mailbox touches a lot of it. The sentence that sells hosting, &ldquo;GDPR-compliant servers in
Germany&rdquo;, answers neither question on its own: the regulation is about who is responsible for
the data and what paper covers each hop, not about the postcode of the rack.</p>
<p>The first half of this page is the four things the GDPR asks of the company that runs an agent
for you, with the articles linked. The second half is a table: every managed OpenClaw host from
<a href="https://aplite.ai/blog/managed-openclaw-hosting-compared">Managed OpenClaw hosting compared</a>, plus the four
European VPS providers the AI assistants name when asked this question, read on their own
privacy, terms and DPA pages on 20 October 2026, with &ldquo;not stated&rdquo; wherever a page is silent.
Aplite is one row in that table, and this page is written by the company behind it.</p>
<h2 id="what-the-gdpr-asks-of-an-agent-host">What the GDPR asks of an agent host</h2>
<h3 id="1-you-are-the-controller-the-host-is-a-processor">1. You are the controller, the host is a processor</h3>
<p>The GDPR assigns roles, not locations. The party that decides what is processed and why is the
<a href="https://gdpr-info.eu/art-4-gdpr/">controller</a>; the party that processes &ldquo;on behalf of the
controller&rdquo; is the processor (Art. 4(7) and 4(8)). When you point an OpenClaw agent at your
notes, your calendar or a mailbox, you are the controller of whatever personal data is in them,
and the company running the gateway for you is your processor. That is the relationship
<a href="https://gdpr-info.eu/art-28-gdpr/">Art. 28(1)</a> regulates: a controller &ldquo;shall use only
processors providing sufficient guarantees&rdquo; of appropriate technical and organisational
measures.</p>
<p>Whether the regulation applies to you at all is decided by <a href="https://gdpr-info.eu/art-3-gdpr/">Art. 3</a>:
by where you are established and whose data it is, &ldquo;regardless of whether the processing
takes place in the Union or not&rdquo;. An EU server does not bring you under the GDPR, and a US
server does not take you out of it.</p>
<p>What the processor holds is not small. An OpenClaw instance keeps its
<a href="https://docs.openclaw.ai/help/faq/where-things-live-on-disk">sessions and transcripts</a> in a
per-agent SQLite database, its memory as markdown files in the workspace, its config and
credentials in <code>~/.openclaw</code>. Every one of those files sits on the host&rsquo;s disk, which is why
the server&rsquo;s location is a real question, just not the only one.</p>
<h3 id="2-a-written-processing-agreement">2. A written processing agreement</h3>
<p><a href="https://gdpr-info.eu/art-28-gdpr/">Art. 28(3)</a> requires that processing by a processor &ldquo;shall
be governed by a contract or other legal act&rdquo; that sets out the subject-matter and duration,
and binds the processor to act only on documented instructions, keep its staff to
confidentiality, take the security measures of Art. 32, help with data-subject requests,
delete or return the data at the end, and allow audits. This is the data processing agreement,
or DPA; German sites call it AVV. A badge that says &ldquo;GDPR compliant&rdquo; is a claim; the DPA is
the document. The table below records whether a host publishes one, offers one on request, or
says nothing.</p>
<h3 id="3-sub-processors-and-the-list-of-them">3. Sub-processors, and the list of them</h3>
<p>A host rarely does everything itself. <a href="https://gdpr-info.eu/art-28-gdpr/">Art. 28(2)</a> says it
&ldquo;shall not engage another processor without prior specific or general written authorisation&rdquo;
and must tell you about changes so you can object; Art. 28(4) passes the same obligations down
to each of those other processors and keeps the first one &ldquo;fully liable to the controller&rdquo;. For
an OpenClaw host the list is predictable: the data centre, the CDN in front of the site, the
mail and payment services, and the one that matters most for an agent, the model provider. A
host that cannot name its sub-processors cannot sign Art. 28 honestly, which is why a published
list is worth more than any badge.</p>
<h3 id="4-where-the-prompts-go">4. Where the prompts go</h3>
<p>The model call leaves the server. Every message you send, every note the agent reads to answer
it and every tool result go to whichever company serves the model, and the frontier models in
October 2026 are served from the United States. That makes the model call a transfer under
Chapter V: <a href="https://gdpr-info.eu/art-44-gdpr/">Art. 44</a> allows it &ldquo;only if&rdquo; the chapter&rsquo;s
conditions are met, through an adequacy decision under Art. 45, for the US the
<a href="https://commission.europa.eu/law/law-topic/data-protection/international-dimension-data-protection/eu-us-data-transfers_en">EU-US Data Privacy Framework</a>
adopted on 10 July 2023 and limited to companies certified under it, or through
<a href="https://gdpr-info.eu/art-46-gdpr/">Art. 46</a> safeguards, in practice the Commission&rsquo;s standard
contractual clauses.</p>
<p>Hosts route that call in one of three ways, and the table&rsquo;s last column records which.</p>
<ul>
<li><strong>Your own key.</strong> The instance calls the provider directly; the host says it is not in the
  path. The provider is then your processor under its own API terms, and the host&rsquo;s DPA does
  not cover it.</li>
<li><strong>The host&rsquo;s account, usually through a router.</strong> The host buys model usage and resells it
  as credits; the router and the providers behind it are the host&rsquo;s sub-processors. The common
  router is OpenRouter, whose <a href="https://openrouter.ai/docs/features/privacy-and-logging">documentation</a>
  says that &ldquo;each AI provider on OpenRouter has its own data handling policies for logging
  and retention&rdquo;, publishes a table of them, and offers EU in-region routing through a
  regional hostname on its Business and Enterprise plans. Whether a host uses that hostname
  is not visible from outside; none of the pages below says it does.</li>
<li><strong>A local model on the box.</strong> Ollama on the same server, and nothing about the conversation
  leaves it. Two hosts in the table describe it; on a VPS you install it yourself.</li>
</ul>
<p>One more thing leaves an OpenClaw server by default, and it is not the prompts: the runtime
makes a <a href="https://docs.openclaw.ai/gateway/telemetry">daily update check</a> to
<code>telemetry.openclaw.ai</code> carrying the version, OS and architecture, with &ldquo;no request body,
install identifier, machine identifier, or random tracking identifier&rdquo;; anonymous feature
statistics are off unless turned on.</p>
<p>So an EU server settles exactly one hop: the data at rest, the transcripts, memory and files,
and the host&rsquo;s own access to them. The model call is the other hop, and it is the same hop on
every host that bills frontier models. What differs between hosts is whether the page says so,
under which paper, and whether a local model is on offer.</p>
<h2 id="the-table">The table</h2>
<p>Managed hosts first, in alphabetical order, then the four VPS providers. &ldquo;Servers&rdquo; is where
the host says your instance runs. &ldquo;Legal entity&rdquo; is the company on the privacy policy, the
terms or the imprint; where only a governing law is given, that is what is written. &ldquo;DPA&rdquo; is
whether a data processing agreement is published or offered, and whether a sub-processor list
exists. &ldquo;Prompts go to&rdquo; is what the host&rsquo;s own pages say about the model call.</p>
<div class="tscroll"><table>
<thead>
<tr>
<th>Host</th>
<th>Servers</th>
<th>Legal entity</th>
<th>DPA / privacy</th>
<th>Prompts go to</th>
</tr>
</thead>
<tbody>
<tr>
<td><a href="https://www.agent37.com/privacy">Agent37</a></td>
<td>not stated; the policy names Cloudflare, Vercel and Daytona, no country</td>
<td>Agent 37 Inc.; New York law; an EU representative (DataRep, Dublin)</td>
<td>privacy policy of 1 Sep 2026; no DPA; SOC 2 Type I; a third-party list, not a sub-processor list</td>
<td>your key: &ldquo;sent from your container straight to the provider&rdquo;; starter credits through Agent37&rsquo;s proxy to OpenRouter</td>
</tr>
<tr>
<td><a href="https://aplite.ai/privacy">Aplite</a></td>
<td>Germany: Nuremberg, netcup; backups on the same server</td>
<td>PIKsolution OÜ, Tallinn, Estonia</td>
<td>privacy policy with a processor table (netcup, Cloudflare, OpenRouter, Polar, Resend, Migadu, Google), SCCs and DPF where the provider holds one; no standalone DPA published</td>
<td>OpenRouter, Inc. (USA), then the provider of the model you picked; standard endpoint, not EU in-region routing; API terms that exclude training</td>
</tr>
<tr>
<td><a href="https://www.ampere.sh/privacy">Ampere</a></td>
<td>Hetzner Cloud, country not stated; account data in Firebase</td>
<td>not stated; a Newark, Delaware address</td>
<td>privacy policy of 10 Feb 2026; no DPA; no transfer mechanism</td>
<td>your key: &ldquo;All LLM API calls go directly from your instance to your chosen provider&rdquo;</td>
</tr>
<tr>
<td><a href="https://www.betterclaw.io/privacy-policy">BetterClaw</a></td>
<td>Railway and Fly.io, United States, per its sub-processor list</td>
<td>BitQit Private Limited, New Delhi, India</td>
<td>privacy policy of Sep 2026, a <a href="https://www.betterclaw.io/legal/dpa">DPA</a> with SCCs 2021/914 and the UK addendum, a <a href="https://www.betterclaw.io/legal/subprocessors">sub-processor list</a></td>
<td>managed models: OpenRouter (USA), then Anthropic, OpenAI, Google, Z.ai or Moonshot; long-term memory processed by OpenAI; your key: your provider, not a sub-processor</td>
</tr>
<tr>
<td><a href="https://clawhosters.com/privacy">ClawHosters</a></td>
<td>Germany: Hetzner, Falkenstein and Nuremberg; &ldquo;All servers are located in German data centers&rdquo;</td>
<td>Daniel Samer, sole proprietor, Duisburg, Germany; &ldquo;a product of yixn.io&rdquo;</td>
<td>privacy policy of Feb 2026 naming Hetzner, Stripe and Coinbase Commerce (SCCs); no DPA; &ldquo;GDPR Compliant&rdquo; badge</td>
<td>included Kimi K2.5, Gemini Flash and DeepSeek: which provider serves them is not stated; your key optional</td>
</tr>
<tr>
<td><a href="https://clawtrust.ai/privacy">ClawTrust</a></td>
<td>United States: &ldquo;dedicated servers in the United States&rdquo;; EU-only processing on request, possibly at a fee</td>
<td>AutoRev AI LLC dba ClawTrust, Columbus, Ohio</td>
<td>privacy policy and a <a href="https://clawtrust.ai/dpa">DPA</a> of 15 Feb 2026 with SCCs Module Two and the UK addendum; sub-processor list on request, 30 days&rsquo; notice of changes</td>
<td>an unnamed &ldquo;AI inference provider&rdquo; for &ldquo;AI model routing&rdquo;, United States; on Lite, your NVIDIA key</td>
</tr>
<tr>
<td><a href="https://everpod.ai/privacy">Everpod</a></td>
<td>Germany or Finland, Hetzner, backups in the same region; account database and website in the United States</td>
<td>DEH Technologies LLC, Wyoming, dba Everpod</td>
<td>privacy policy of 18 Oct 2026 with a provider table (Hetzner, Cloudflare, Supabase, Render, OpenRouter, Postmark, PostHog and more); relies on its providers&rsquo; DPAs and SCCs; no DPA of its own stated</td>
<td>included usage: OpenRouter under Everpod&rsquo;s account, to OpenAI, Anthropic, Google, xAI and others, &ldquo;only to providers that don&rsquo;t train on it&rdquo;; your key: your own agreement</td>
</tr>
<tr>
<td><a href="https://runonflux.com/privacy-policy/">Flux</a></td>
<td>&ldquo;FluxNodes operate in 67 countries&rdquo;; personal data &ldquo;to Switzerland&rdquo; for processing; no way to pick a node described</td>
<td>InFlux Technologies; Swiss law; address not stated</td>
<td>privacy policy without a date; no DPA; no transfer mechanism named</td>
<td>your key: &ldquo;bring your own API keys&rdquo;</td>
</tr>
<tr>
<td><a href="https://www.hostinger.com/legal/privacy-policy">Hostinger</a></td>
<td>picked at order: France, Germany, Lithuania, UK; India, Indonesia, Malaysia; USA (Phoenix, Boston); Brazil</td>
<td>HOSTINGER, UAB and HOSTINGER operations, UAB, Vilnius, Lithuania; the DPA is signed by Hostinger International Ltd., Cyprus</td>
<td>privacy policy of 14 Oct 2026, a <a href="https://www.hostinger.com/legal/dpa">DPA</a> with SCCs Module Two and Three, the UK addendum and eight sub-processors in its appendix (AWS EMEA, Google Cloud EMEA, Cloudflare, Anthropic Ireland among them)</td>
<td>&ldquo;Hostinger AI Router&rdquo;: Claude, GPT and Gemini &ldquo;through a single managed integration&rdquo;; or a ChatGPT subscription; or your key for OpenAI, Anthropic, xAI, Gemini</td>
</tr>
<tr>
<td>KiloClaw</td>
<td>—</td>
<td>—</td>
<td>—</td>
<td>—</td>
</tr>
<tr>
<td><a href="https://myclaw.ai/privacy">MyClaw</a></td>
<td>not stated; data &ldquo;may be transferred to … countries outside your jurisdiction, including the United States&rdquo;</td>
<td>not stated; Delaware law</td>
<td>privacy policy of 11 Sep 2026; no DPA; &ldquo;Standard Contractual Clauses or other lawful transfer mechanisms&rdquo;</td>
<td>&ldquo;the AI service provider selected or configured by the user&rdquo;; tokens not included</td>
</tr>
<tr>
<td><a href="https://oneclickclaw.io/privacy">OneClickClaw</a></td>
<td>Denmark: Webdock; backups in the same EU region</td>
<td>sole proprietorship of Luigi Ramos, Greece</td>
<td>privacy policy of 30 Sep 2026 and a <a href="https://oneclickclaw.io/legal/dpa">DPA</a> with a sub-processor table (Webdock, Denmark; Stripe, Google, Resend, Cloudflare, Anthropic, USA, under SCCs), labelled &ldquo;Draft pending legal review&rdquo;</td>
<td>your key or a ChatGPT / SuperGrok sign-in: your provider, &ldquo;the Customer&rsquo;s own sub-processor&rdquo;; Anthropic only for OneClickClaw&rsquo;s support chatbot</td>
</tr>
<tr>
<td><a href="https://setupopenclaw.com/privacy">OpenClaw Cloud</a> (setupopenclaw.com)</td>
<td>Germany: &ldquo;All servers run in Hetzner&rsquo;s German data centers&rdquo;</td>
<td>not stated; the policy says &ldquo;provided by the OpenClaw team&rdquo;, which the project&rsquo;s own site contradicts</td>
<td>privacy policy of Jan 2026; no DPA; no model provider named</td>
<td>credits sold separately or your key; the provider behind the credits is not named</td>
</tr>
<tr>
<td><a href="https://openclawlaunch.com/privacy">OpenClaw Launch</a></td>
<td>&ldquo;servers located in the United States and the European Union&rdquo;; Hetzner and Netcup named, which is where not stated</td>
<td>not stated; Delaware law</td>
<td>privacy policy of 16 Oct 2026; a written DPA &ldquo;contact us&rdquo; on request; sub-processors named inline (Hetzner, Netcup, Cloudflare, OpenRouter)</td>
<td>OpenRouter, then OpenAI, Anthropic, Google or DeepSeek; or your own OpenRouter key</td>
</tr>
<tr>
<td><a href="https://railway.com/legal/privacy">Railway</a></td>
<td>picked per service: Amsterdam, US West (California), US East (Virginia), Singapore; &ldquo;primary processing takes place in the United States&rdquo;</td>
<td>Railway Corporation, San Francisco</td>
<td>privacy policy, a <a href="https://railway.com/legal/dpa">DPA</a> with DPF, SCCs 2021/914 and the UK addendum, a <a href="https://trust.railway.com/item/subprocessors">sub-processor list</a></td>
<td>your key: the template takes only the gateway token; your provider</td>
</tr>
<tr>
<td><a href="https://rapidclaw.dev/privacy">RapidClaw</a></td>
<td>not stated; Vercel for hosting, &ldquo;cloud infrastructure providers&rdquo; unnamed</td>
<td>&ldquo;Rapid Claw&rdquo;; entity, address and jurisdiction not stated</td>
<td>privacy policy of 10 Mar 2026; no DPA; no transfer mechanism</td>
<td>credits included; the provider is not named</td>
</tr>
<tr>
<td><a href="https://runware.ai/privacy">Runware</a></td>
<td>&ldquo;Data residency EU &amp; US&rdquo; in the footer; which one an instance uses is not stated</td>
<td>Runware Ltd, Harrow, United Kingdom; the footer says Runware, Inc.</td>
<td>privacy policy without a date; no DPA; GDPR, ISO 27001 and SOC 2 badges</td>
<td>open-source models on Runware&rsquo;s own infrastructure; closed-source models &ldquo;pass-through&rdquo;, upstream not named</td>
</tr>
<tr>
<td><a href="https://xcloud.host/privacy-policy/">xCloud</a></td>
<td>picked from <a href="https://xcloud.host/docs/31-server-locations-worldwide/">31 locations</a>, six in the EU: Frankfurt, Amsterdam, Paris, Madrid, Stockholm, Warsaw</td>
<td>xCloud Hosting, LLC, Milton, Delaware</td>
<td>privacy policy effective 10 Sep 2026; no DPA; no transfer mechanism</td>
<td>your key: Anthropic, OpenAI, OpenRouter, Moonshot or Gemini; &ldquo;we do not receive your conversations&rdquo;</td>
</tr>
<tr>
<td><a href="https://docs.hetzner.com/general/general-terms-and-conditions/data-privacy-faq">Hetzner</a> (VPS)</td>
<td>Nuremberg and Falkenstein, Germany; Helsinki, Finland; cloud also in Ashburn and Hillsboro, USA, and Singapore; &ldquo;the storage location depends on the product location you have chosen&rdquo;</td>
<td>Hetzner Online GmbH, Gunzenhausen, Germany, HRB 6089</td>
<td>privacy policy of 14 Sep 2026; DPA concluded by checkbox in the customer account, sample contract as PDF</td>
<td>whichever provider you configure; a local model keeps it on the box</td>
</tr>
<tr>
<td><a href="https://www.ovhcloud.com/en-ie/terms-and-conditions/contracts/">OVHcloud</a> (VPS)</td>
<td>&ldquo;46 datacentres on 4 continents&rdquo;, region picked per service</td>
<td>OVH SAS, Roubaix, France</td>
<td>a Data Processing Agreement and an OVH Sub-processors list as PDFs on the contracts page; ISO 27001, 27701, SOC, SecNumCloud</td>
<td>whichever provider you configure</td>
</tr>
<tr>
<td><a href="https://www.scaleway.com/en/privacy-policy/">Scaleway</a> (VPS)</td>
<td>Paris, Amsterdam, Warsaw, Milan, three availability zones each</td>
<td>Scaleway SAS, Paris</td>
<td>privacy policy of June 2026; a <a href="https://www.scaleway.com/en/contracts/">DPA</a> as PDF; SCCs 2021/914 for transfers</td>
<td>whichever provider you configure; Scaleway&rsquo;s own Generative APIs are &ldquo;All hosted in Europe&rdquo;</td>
</tr>
<tr>
<td><a href="https://www.netcup.com/en/contact/data-privacy">netcup</a> (VPS)</td>
<td>picked at order: Nuremberg, Vienna, Amsterdam, or &ldquo;No preference Europe&rdquo;; also Manassas, USA, and Singapore</td>
<td>netcup GmbH, Karlsruhe, Germany, HRB 705547</td>
<td>privacy policy effective 5 Aug 2026; ISO 27001 (TÜV Nord) and ISO 27701; a customer DPA was not found on the public pages</td>
<td>whichever provider you configure</td>
</tr>
</tbody>
</table></div>
<h2 id="reading-the-table">Reading the table</h2>
<p><strong>Servers in the EU.</strong> Five managed hosts put the instance itself in the EU without being
asked: Aplite (Germany), ClawHosters (Germany), Everpod (Germany or Finland), OneClickClaw
(Denmark) and OpenClaw Cloud (Germany, Hetzner). Three let you pick an EU location at order
time: Hostinger (France, Germany, Lithuania), xCloud (six EU cities) and Railway (Amsterdam).
Two say &ldquo;EU and US&rdquo; without saying which you get: OpenClaw Launch and Runware. ClawTrust and
BetterClaw run in the United States; Flux processes personal data in Switzerland and runs
containers on nodes in 67 countries. Four say nothing about where the instance runs: Agent37,
Ampere (Hetzner, no country), MyClaw and RapidClaw. All four VPS providers let you choose an
EU data centre.</p>
<p><strong>An EU legal entity.</strong> The company you would sign with is in the EU at Aplite (Estonia),
ClawHosters (Germany), OneClickClaw (Greece) and Hostinger (Lithuania, with a Cypriot entity
on the DPA), and at all four VPS providers. Runware is a UK company. Agent37, ClawTrust,
Everpod, Railway and xCloud are US companies; MyClaw and OpenClaw Launch name only Delaware
law; Ampere names only a Delaware address; BetterClaw is an Indian company; Flux names Swiss
law. RapidClaw and OpenClaw Cloud name no entity at all.</p>
<p><strong>A DPA you can read.</strong> Published: BetterClaw, ClawTrust, Hostinger, Railway, and
OneClickClaw with a draft notice; on request: OpenClaw Launch. Among the VPS providers,
Hetzner, OVHcloud and Scaleway publish one; netcup&rsquo;s was not found. The other eleven managed
hosts, Aplite among them, have a privacy policy and no separate agreement. A sub-processor
list exists at BetterClaw, Hostinger and OneClickClaw (inside the DPA), Railway (a trust
page), OVHcloud (a PDF), and as a processor table in the privacy policies of Aplite and
Everpod; ClawTrust gives it on request.</p>
<p><strong>Where the prompts go.</strong> Through OpenRouter on the host&rsquo;s account: Aplite, BetterClaw,
Everpod, OpenClaw Launch, and Agent37 for its starter credits. Through the host&rsquo;s own routing:
Hostinger (its AI Router, with Anthropic Ireland on the sub-processor list) and Runware (its
own inference for open-source models). Through a provider the page does not name: ClawHosters,
ClawTrust, RapidClaw, OpenClaw Cloud. Only your own key: Ampere, Flux, MyClaw, OneClickClaw,
Railway, xCloud, and every VPS. OpenRouter&rsquo;s EU in-region routing is mentioned by none of them.</p>
<h2 id="if-the-prompts-have-to-stay-in-the-eu-too">If the prompts have to stay in the EU too</h2>
<p>Three ways, from the table and the upstream docs.</p>
<ul>
<li><strong>A local model on the server.</strong> OneClickClaw runs Ollama on its Hobbyist and Pro plans;
  ClawHosters connects Ollama on your own hardware over a VPN; on any VPS you install
  <a href="https://docs.openclaw.ai/gateway/local-models">Ollama</a> yourself. Nothing about the
  conversation leaves the machine, and the models are the smaller open-weight ones. The $0 path
  is in <a href="https://aplite.ai/blog/openclaw-hosting-for-free">OpenClaw hosting for free</a> and the key question in
  <a href="https://aplite.ai/blog/openclaw-without-api-key">Can you use OpenClaw without an API key?</a>.</li>
<li><strong>An EU-hosted model API with your own key.</strong> Scaleway&rsquo;s Generative APIs are &ldquo;All hosted in
  Europe&rdquo;; a BYOK host or a VPS lets you point OpenClaw at them. The provider is then your
  processor under its own terms.</li>
<li><strong>A router&rsquo;s EU endpoint.</strong> OpenRouter offers in-region EU routing on Business and
  Enterprise plans; a BYOK host or a VPS lets you use your own account with it. None of the
  managed hosts that resell OpenRouter say they use it, Aplite included.</li>
</ul>
<h2 id="the-aplite-row-in-full">The Aplite row, in full</h2>
<p>Aplite runs one OpenClaw instance per workspace in its own container on a server at netcup
in Nuremberg, Germany; the database and the encrypted backups are on the same server. The
company is PIKsolution OÜ, Tallinn, Estonia, named as the controller in the
<a href="https://aplite.ai/privacy">privacy policy</a> and the <a href="https://aplite.ai/impressum">imprint</a>. The policy lists every processor
with its country: netcup (Germany), Cloudflare (USA, in front of every request), OpenRouter
and the model providers behind it (USA), Polar (EU and USA, payments), Resend (USA, email),
Migadu (France, mailboxes), Google (EU and USA, sign-in only). For transfers outside the EU it
relies on the standard contractual clauses and, where the provider holds one, a Data Privacy
Framework certification.</p>
<p>The model call: the agent&rsquo;s messages, the notes it reads and the tool results go to
OpenRouter&rsquo;s standard endpoint and on to the provider of the model you picked from the catalog,
under API terms that exclude training. Aplite does not use OpenRouter&rsquo;s EU in-region routing,
does not offer local models, and has no option to bring your own key; what the agent sends is
limited to the request at hand, and the policy says so. There is no standalone data processing
agreement on the site; the privacy policy is the public statement of how the data is handled,
and questions go to support@aplite.ai. None of this is a claim that your deployment is
compliant; that depends on what you put into it and on the paper you hold with each party.
The rest of what Aplite does and does not do with OpenClaw is on the
<a href="https://aplite.ai/docs/openclaw">OpenClaw page</a>.</p>
<p>Every cell in the table was read on 20 October 2026; legal pages change, and the plan is to
re-read them once a quarter alongside the <a href="https://aplite.ai/blog/managed-openclaw-hosting-compared">prices</a>.</p>
<h2 id="faq">FAQ</h2>
<h3 id="is-openclaw-gdpr-compliant">Is OpenClaw GDPR compliant?</h3>
<p>OpenClaw is software you run, and compliance is a property of a deployment, not of a program.
What the software does is documented: it keeps sessions, memory, config and credentials on
the machine it runs on, sends the conversation to whichever model provider you configure, and
makes a daily update check to the project&rsquo;s server that carries no prompts or identifiers.
Whether a deployment meets the GDPR depends on who runs that machine for you and under which
agreement, which provider the prompts go to and on what terms, and whether you have a lawful
basis for the data you feed in.</p>
<h3 id="does-an-eu-server-make-my-agent-gdpr-compliant">Does an EU server make my agent GDPR-compliant?</h3>
<p>No. An EU server settles where the data at rest sits, the transcripts, memory and files, and
which law the host answers to. It does not settle the model call, which goes to the provider of
the model, in October 2026 usually in the United States, and it does not replace the written
processing agreement that Art. 28 requires between you and the host. A server in Frankfurt
with no DPA and an unnamed model provider is a server, not compliance.</p>
<h3 id="which-managed-openclaw-hosts-are-in-the-eu">Which managed OpenClaw hosts are in the EU?</h3>
<p>Hosts whose pages put the instance in the EU: Aplite (Germany, Estonian company), ClawHosters
(Germany, German sole proprietor), Everpod (Germany or Finland, US company), OneClickClaw
(Denmark, Greek sole proprietor), OpenClaw Cloud (Germany, Hetzner, no entity named). Hosts
where you pick an EU location: Hostinger, xCloud and Railway. OpenClaw Launch and Runware say
&ldquo;EU and US&rdquo; without saying which. Hetzner, OVHcloud, Scaleway and netcup rent EU servers for
self-hosting.</p>
<h3 id="which-openclaw-hosts-offer-a-data-processing-agreement">Which OpenClaw hosts offer a data processing agreement?</h3>
<p>Published on the site: BetterClaw, ClawTrust, Hostinger, Railway, and OneClickClaw with a
&ldquo;draft pending legal review&rdquo; label. On request: OpenClaw Launch. Among VPS providers: Hetzner
(in the customer account), OVHcloud and Scaleway (PDF). The others, Aplite included, publish
a privacy policy and no separate agreement as of 20 October 2026.</p>
<h3 id="where-do-my-prompts-go-on-a-managed-openclaw-host">Where do my prompts go on a managed OpenClaw host?</h3>
<p>To the company that serves the model. With your own key, straight from the instance to that
provider under your agreement with it. With the host&rsquo;s credits, through the host&rsquo;s account,
most often via OpenRouter in the United States and then to OpenAI, Anthropic, Google or the
open-weight provider behind the model; each of those has its own retention policy, which
OpenRouter documents per provider. With a local model, nowhere: the conversation stays on the
server. The table&rsquo;s last column says which of the three each host describes.</p>]]></content:encoded>
  </item>
  <item>
    <title>Managed OpenClaw vs self-hosting: what you give up and what you keep</title>
    <link>https://aplite.ai/blog/managed-openclaw-vs-self-hosting</link>
    <guid isPermaLink="true">https://aplite.ai/blog/managed-openclaw-vs-self-hosting</guid>
    <pubDate>Tue, 20 Oct 2026 00:00:00 +0000</pubDate>
    <description>Twelve factors side by side for a VPS, the machine on your desk and a managed host: setup, updates, root and shell, openclaw.json, skills and plugins, models and keys, backups, export, security, uptime, price and your time. The self-hosted columns from the OpenClaw docs, the managed column from each vendor&#x27;s own page, Aplite&#x27;s limits from its docs and code. Checked 20 October 2026.</description>
    <content:encoded><![CDATA[<figure class="fig">
<a class="zoom" href="https://aplite.ai/assets/blog/managed-openclaw-vs-self-hosting.webp?v=98d5c29f" target="_blank" rel="noopener"><img src="https://aplite.ai/assets/blog/managed-openclaw-vs-self-hosting.webp?v=98d5c29f" alt="One gateway, three operators: a rented server, the machine on your desk, a managed host — and the rows where they differ." width="1560" height="820" decoding="async"></a>
<figcaption>One gateway, three operators: a rented server, the machine on your desk, a managed host — and the rows where they differ.</figcaption>
</figure>
<p>Ask an AI assistant whether to put OpenClaw on a Hostinger VPS or on a managed host and
the answer comes back as a table: setup, maintenance, root access, backups, portability,
each cell a phrase like &ldquo;often limited&rdquo; or &ldquo;depends on provider&rdquo;. The phrases are true and
do not help, because what a managed host keeps for itself is specific to the host, and
what self-hosting hands you is specific too: a config file at a known path, a backup
command, an update you run yourself, a security checklist that is yours to work through.</p>
<p>This page fills the table in. The two self-hosted columns are read from the
<a href="https://docs.openclaw.ai">OpenClaw documentation</a> on 20 October 2026, when the current
stable release was 2026.10.5. The managed column says only what a host&rsquo;s own page says,
drawn from the eighteen services in
<a href="https://aplite.ai/blog/managed-openclaw-hosting-compared">Managed OpenClaw hosting compared</a>. Aplite,
which runs this site, gets a column of its own, and its limits are taken from its
<a href="https://aplite.ai/docs/openclaw">OpenClaw page</a> and from its code rather than from its pitch. The runtime
is the same MIT-licensed OpenClaw in every column; the question is who operates it and
what the operator keeps.</p>
<h2 id="the-table">The table</h2>
<div class="tscroll"><table>
<thead>
<tr>
<th>Factor</th>
<th>Self-hosted: a VPS</th>
<th>Self-hosted: your own machine</th>
<th>Managed hosts</th>
<th>Aplite</th>
</tr>
</thead>
<tbody>
<tr>
<td>Setup</td>
<td>Rent the server, SSH in, run the one-line installer, answer onboarding, install the Gateway as a systemd service; Node 24.16 or newer</td>
<td>The same installer on macOS, Linux or Windows; a LaunchAgent, systemd user service or Scheduled Task starts it at login</td>
<td>An account and a plan; &ldquo;one-click&rdquo; at Hostinger, a Google account at OneClickClaw; hosts that use your API key ask for it here</td>
<td>From the browser, under a minute; no key, no pairing by you</td>
</tr>
<tr>
<td>Updates</td>
<td><code>openclaw update</code>, or <code>/update</code> from a chat; the docs say back up first; three stable releases in five days this October</td>
<td>The same</td>
<td>The host&rsquo;s schedule: OneClickClaw ships &ldquo;versions we have tested ourselves&rdquo; and lets you pin one; xCloud&rsquo;s &ldquo;apply themselves&rdquo;; Hostinger lists &ldquo;one-click updates&rdquo;</td>
<td>Our schedule, after testing: new workspaces run 2026.10.10 since 3 November, the upstream stable of that day; on 20 October the fleet was on 2026.9.8 against upstream&rsquo;s 2026.10.5</td>
</tr>
<tr>
<td>Root and shell</td>
<td>Root</td>
<td>Your user, your disk</td>
<td>From full SSH (OneClickClaw) and a web terminal (Agent37) to &ldquo;no Docker, no terminal, no SSH&rdquo; (xCloud); not stated at Hostinger, Everpod</td>
<td>None for you; the agent runs commands inside its own container</td>
</tr>
<tr>
<td><code>openclaw.json</code></td>
<td><code>~/.openclaw/openclaw.json</code>, JSON5, every section yours</td>
<td>The same</td>
<td>Reachable where there is a shell; not mentioned on the other pages</td>
<td>Not exposed: agents, models, skills and channels are set in the browser and we write the file; what the UI does not show cannot be changed</td>
</tr>
<tr>
<td>Skills and plugins</td>
<td><code>openclaw skills install</code> from ClawHub, git or a folder; <code>openclaw plugins install</code> from ClawHub, npm or git</td>
<td>The same</td>
<td>Rarely stated; OneClickClaw: &ldquo;Anything OpenClaw can do, your server can do&rdquo;</td>
<td>A curated catalog, 24 skills today, switched on in the browser; agents write their own; your MCP servers connect; on Pro and Power the agent installs packages inside its own container, on Free it does not; no arbitrary plugins</td>
</tr>
<tr>
<td>Models and keys</td>
<td>Any bundled provider or a custom one, a ChatGPT or Codex login, or a local runtime (Ollama, LM Studio, llama.cpp, vLLM); the keys are yours</td>
<td>The same, and the local model runs on the hardware in front of you</td>
<td>Three groups: the host&rsquo;s account, a subscription sign-in, or your key; local models at OneClickClaw (Ollama on its servers) and ClawHosters (yours, over a VPN)</td>
<td>21 catalog models through one metered account, billed in credits; no own key, no local models; Free gets the budget end</td>
</tr>
<tr>
<td>Backups</td>
<td><code>openclaw backup create</code> by hand, or <code>backup enable</code> for a daily job: state, config, credentials, agents, workspaces; storing it off the machine, encrypted, is yours</td>
<td>The same</td>
<td>Daily at xCloud, Everpod, OpenClaw Launch, MyClaw and RapidClaw; daily and weekly snapshots at OneClickClaw; an add-on at ClawHosters; yours at Railway; not stated at Runware</td>
<td>A nightly encrypted backup of every workspace, kept 7 daily, 4 weekly and 3 monthly; no backup button and no self-serve restore</td>
</tr>
<tr>
<td>Export and portability</td>
<td>Everything is on your disk; the backup archive restores on another machine</td>
<td>The same</td>
<td>Everpod: a .zip on request; Agent37: &ldquo;plain files you can download&rdquo;; OneClickClaw: &ldquo;copy it off and go&rdquo;; Hostinger: settings &ldquo;don&rsquo;t transfer automatically&rdquo;; xCloud: not stated</td>
<td>Notes and agent instructions are plain <code>.md</code> and download one by one; no archive of the runtime state; leaving means a fresh install plus your files</td>
</tr>
<tr>
<td>Security: gateway and approvals</td>
<td>The Gateway binds to loopback; exposing it (firewall, Tailscale, reverse proxy) is yours; host commands run with no approval prompt by default; sandboxing off by default</td>
<td>The same, and the agent runs as your user on your files</td>
<td>The host&rsquo;s machine and firewall; rarely detailed; OneClickClaw: &ldquo;a machine nobody else shares&rdquo;</td>
<td>One container per workspace with its own network; the gateway port bound to localhost on our host; pairing done by us; every file change and outbound action waits for your approval</td>
</tr>
<tr>
<td>Always on</td>
<td>Yes, for the rent</td>
<td>Only while awake and online</td>
<td>Paid plans yes; free tiers sleep (Agent37 after 15 minutes)</td>
<td>Pro and Power yes; Free sleeps after 6 hours idle and wakes on the next visit</td>
</tr>
<tr>
<td>Price</td>
<td>$5 to $12 a month plus the models, about $1.50 to $31 at thirty messages a day</td>
<td>Electricity, under $1 to $9 a month for a Mac mini, plus the models, or $0 with Ollama</td>
<td>$0 to $29 for the entry plan; some include model usage</td>
<td>Free $0 with 300 credits; Pro $20 with 2,000 a month; Power $50; 1 credit = 1¢ of model use</td>
</tr>
<tr>
<td>Your time</td>
<td>Install, updates, backups, the night it stops</td>
<td>The same, plus a machine that must stay awake</td>
<td>Agents and channels; the box is the host&rsquo;s</td>
<td>Agents, channels and approvals</td>
</tr>
</tbody>
</table></div>
<p>The rest of the page is the table row by row, with the quotes and the links, then what
a managed instance keeps for itself in general and what Aplite keeps in particular, how
to move between the columns, and which column fits which requirement.</p>
<h2 id="setup">Setup</h2>
<p>Self-hosting starts with the <a href="https://docs.openclaw.ai/install">install page</a>: Node
24.16 or newer, <code>curl -fsSL https://openclaw.ai/install.sh | bash</code> on macOS, Linux or
WSL2, <code>iwr -useb https://openclaw.ai/install.ps1 | iex</code> on Windows, then onboarding, and
<code>openclaw gateway install</code> for a LaunchAgent on macOS, a systemd user service on Linux
or a Scheduled Task on Windows so the Gateway comes back after a reboot. On a VPS the
same steps follow a login over SSH, and the docs have provider pages for
<a href="https://docs.openclaw.ai/install/hetzner">Hetzner</a> and
<a href="https://docs.openclaw.ai/install/digitalocean">DigitalOcean</a> among others. Onboarding
asks for the model provider and a key, which is the step a managed host with its own
model account removes.</p>
<p>A managed host starts with an account. Hostinger sells &ldquo;one-click&rdquo; deployment on a VPS
plan; OneClickClaw takes a Google account for its seven-day trial; the hosts in the
&ldquo;your API key&rdquo; group of the comparison ask for the key during setup. On Aplite the setup
is the browser: pick a plan and a team, press Deploy, and the instance is online in under
a minute with the model account already attached and the device pairing already done.</p>
<h2 id="updates">Updates</h2>
<p>OpenClaw ships often. The <a href="https://github.com/openclaw/openclaw/releases">releases feed</a>
showed three stable versions in five days this month: 2026.9.33 on 15 October, 2026.10.4
on 16 October and 2026.10.5 on 19 October, with 2026.10.10 in beta the same day.
Self-hosted, each one is your decision. The <a href="https://docs.openclaw.ai/install/updating">updating page</a>
describes <code>openclaw update</code>: it &ldquo;detects your install type (npm, pnpm, Bun, or git)&rdquo;,
checks the new version &ldquo;while the old Gateway serves, then activates and verifies the
update&rdquo;, and can be triggered as <code>/update</code> from a connected chat. The same page says
&ldquo;Before a significant update, create a verified backup&rdquo; and that &ldquo;Automatic config copies
and migration recovery originals are not a full-state backup&rdquo;; installs older than July
2026 must step through 2026.9.5 first.</p>
<p>Managed, the schedule is the host&rsquo;s. OneClickClaw: &ldquo;We auto-update OpenClaw to the stable
version we trust in production&rdquo;, &ldquo;You get the OpenClaw versions we have tested
ourselves&rdquo;, and &ldquo;You can pin any version from your dashboard&rdquo;. xCloud: &ldquo;Updates that
apply themselves&rdquo;. Hostinger&rsquo;s application page lists &ldquo;One-click updates&rdquo; without saying
who clicks. Everpod: &ldquo;We keep your agent&rsquo;s software current and roll changes out
carefully&rdquo;. On Aplite the fleet moves when we have tested a release, and the gap is a
number you can check: on 20 October 2026 every Aplite workspace ran 2026.9.8 while the
upstream stable was 2026.10.5; on 3 November new workspaces moved to 2026.10.10, released
that day, and existing ones follow on their next start. New features arrive days or
weeks, not hours, after they ship.</p>
<h2 id="root-and-shell">Root and shell</h2>
<p>On a VPS you are root; on your own machine the Gateway runs as your user with your
user&rsquo;s reach. Among managed hosts the row runs the whole range. OneClickClaw: &ldquo;You get
full SSH with admin rights on a machine nobody else shares&rdquo;, opt-in from the dashboard,
free on every tier. Agent37 puts &ldquo;a task board, web terminal, file browser, live desktop,
and scheduled jobs, all in your browser&rdquo; and says &ldquo;Your files and keys live inside your
container&rdquo;. xCloud describes its service as &ldquo;no Docker, no terminal, no SSH&rdquo;. Hostinger&rsquo;s
OpenClaw pages mention Docker container management and container logs and do not mention
a shell; Everpod&rsquo;s page does not either.</p>
<p>Aplite gives you no shell. The agent has one: <code>exec</code> is in its tool list, and it runs
commands inside its own container, the same place its files live. Every file change it
makes, and every action that leaves the workspace, arrives as a proposal in
<a href="https://aplite.ai/docs/approvals">Approvals</a> before it happens. So there is a terminal in the picture, but
it is the agent&rsquo;s, not yours, and you see its results rather than typing into it.</p>
<h2 id="openclawjson">openclaw.json</h2>
<p>Self-hosted, the whole runtime is one file. The
<a href="https://docs.openclaw.ai/gateway/configuration-reference">configuration reference</a> is a
&ldquo;field-level reference for <code>~/.openclaw/openclaw.json</code>&rdquo;, in JSON5 &ldquo;(comments + trailing
commas allowed)&rdquo;, with sections for channels, agents, tools, models, MCP, skills, plugins,
gateway, hooks, automations (<code>cron</code>), update and the rest. Anything the runtime can do
has a key here.</p>
<p>Managed hosts&rsquo; pages do not talk about the file. Where a host gives you a shell, the file
is on a machine you can log into; where it does not, the page is silent on whether you
can read it. On Aplite the file exists, one per workspace, and you do not see it: agents,
their models, skills and channels are configured in the browser, and the platform
regenerates <code>openclaw.json</code> from those settings on every change. The consequence to be
clear about is the inverse: a setting the interface does not expose cannot be changed.
The heartbeat interval, the tool allowlist, the sandbox policy and the provider routing
are set by us for every workspace.</p>
<h2 id="skills-and-plugins">Skills and plugins</h2>
<p>Self-hosted, the <a href="https://docs.openclaw.ai/cli/skills">skills CLI</a> installs from ClawHub
(<code>openclaw skills install @owner/&lt;slug&gt;</code>), from git or from a local folder, and the
<a href="https://docs.openclaw.ai/cli/plugins/install">plugins CLI</a> takes <code>clawhub:</code>, <code>npm:</code> and
<code>git:</code> sources, so anything published anywhere is a command away. Managed pages mostly
say nothing about either. OneClickClaw&rsquo;s &ldquo;Anything OpenClaw can do, your server can do&rdquo;
is the broadest statement, and with SSH it holds. Everpod mentions skills only in its
export FAQ, &ldquo;the skills it made&rdquo;.</p>
<p>On Aplite skills come from a <a href="https://aplite.ai/docs/skills-and-connections">curated catalog</a>, 24 of them
switched on today, enabled per agent in the browser; agents also write their own skills
into their own folder, and those survive a resync. Servers you run yourself connect as
MCP servers from the Skills page on every plan. A set of command-line tools (ffmpeg,
pandoc, jq, ripgrep, the GitHub CLI, DuckDB and others) is switched on per workspace from
the Server page. On Free the agent installs nothing; on Pro and Power it can install
packages (npm, pip, apt) inside its own container. OpenClaw plugins are not installable on
any plan; a skill that needs one is not available until we add it.</p>
<h2 id="models-and-keys">Models and keys</h2>
<p>The <a href="https://docs.openclaw.ai/concepts/model-providers">model providers</a> page lists the
bundled providers (OpenAI, Anthropic, Google Gemini and Vertex, OpenRouter, xAI, Z.AI,
MiniMax, NVIDIA, Kilo Gateway, Vercel AI Gateway, Moonshot and more), the ChatGPT and
Codex OAuth login, custom providers under <code>models.providers</code>, and local runtimes: Ollama,
LM Studio, llama.cpp, vLLM, SGLang. Self-hosted, the key is yours and the bill is the
provider&rsquo;s; on your own machine a local model makes the model line $0, with the caveats
the upstream FAQ attaches to small models.</p>
<p>Managed hosts fall into the three groups the comparison already sorted. The host runs the
model account: Aplite, Runware, Hostinger, Everpod, RapidClaw, ClawHosters, OpenClaw
Launch, Ampere. A subscription sign-in: OneClickClaw (ChatGPT or Grok) and Hostinger
(ChatGPT Plus or Pro). Your own key: xCloud, Flux, Railway, MyClaw, OpenClaw Cloud,
BetterClaw Basic. Local models on a managed host appear on two pages: OneClickClaw
installs Ollama on its Hobbyist and Pro servers, ClawHosters connects Ollama on your own
hardware over a VPN. Whether a key can be avoided at all is the subject of
<a href="https://aplite.ai/blog/openclaw-without-api-key">Can you use OpenClaw without an API key?</a>.</p>
<p>Aplite&rsquo;s column is the narrowest here, on purpose: 21 models in the
<a href="https://aplite.ai/docs/models">catalog</a>, routed through one metered OpenRouter account, billed in
credits at 1 credit = 1¢; Free gets the budget end of the catalog, paid plans all of it.
Bringing your own key is not supported, a local model is not an option, and a model that
is not in the catalog is not reachable. The price of a model in credits sits above
OpenRouter&rsquo;s list price; how that works out per message and per month is in
<a href="https://aplite.ai/blog/openclaw-cost-per-month">What OpenClaw really costs per month</a>.</p>
<h2 id="backups">Backups</h2>
<p>Self-hosted, the <a href="https://docs.openclaw.ai/install/backups">backups page</a> gives you
<code>openclaw backup create</code>: a verified <code>.tar.gz</code> of &ldquo;state, config, credentials, every
configured agent directory&rdquo; and, by default, the workspaces, with the SQLite databases
captured through the online backup API. It is manual unless you run
<code>openclaw backup enable</code> for a daily Gateway-owned job; the docs call a daily archive &ldquo;a
reasonable daily routine for small installs&rdquo;, say to run one before updates, resets and
machine moves, and warn that archives &ldquo;contain credentials and session history&rdquo; and
belong encrypted and off the machine. Restore is <code>openclaw backup restore &lt;archive&gt;
--target &lt;new-dir&gt;</code> and &ldquo;never activates in place&rdquo;. On a VPS the provider&rsquo;s snapshots sit
underneath; on your own machine, your own habits do.</p>
<p>Managed, the row is what the page says. Daily: xCloud (&ldquo;daily encrypted backups&rdquo;), Everpod
(&ldquo;daily automatic backups&rdquo; and &ldquo;restores we actually test&rdquo;), OpenClaw Launch, MyClaw,
RapidClaw. OneClickClaw: &ldquo;Daily and weekly snapshots, completely free&rdquo;, a snapshot before
every update, &ldquo;restored in minutes&rdquo;. Hostinger: backups are in &ldquo;we handle the setup,
updates, backups, and security&rdquo;, frequency not stated on the OpenClaw page. ClawHosters:
an add-on on the pricing page. Railway: yours. Runware: not stated.</p>
<p>Aplite backs up every workspace nightly: the database, the runtime state and the
workspace files, in one encrypted bundle, kept for 7 days plus the newest of each of the
last 4 weeks and 3 months. There is no backup button and no self-serve restore; a restore
goes through us, not through a dashboard. That is the row where &ldquo;managed&rdquo; is most
literal.</p>
<h2 id="export-and-portability">Export and portability</h2>
<p>Self-hosted, portability is the same archive: the
<a href="https://docs.openclaw.ai/install/migrating">migrating guide</a> is &ldquo;stop the Gateway, back
up the whole <code>~/.openclaw/</code> directory, not just <code>openclaw.json</code>&rdquo;, move the archive &ldquo;via
scp, an external drive, or another protected channel&rdquo;, restore to a staging directory on
the new machine, run <code>openclaw doctor</code> and restart. Plugin <code>node_modules</code> are reinstalled,
WhatsApp may need relinking, approvals and delivery state roll back to the archive&rsquo;s
moment.</p>
<p>Managed hosts answer in their own words. Everpod: &ldquo;ask it to pack itself into one .zip
file&rdquo;, and on leaving &ldquo;we&rsquo;ll send you a full copy of your pod&rsquo;s data&rdquo;. Agent37: &ldquo;your
workspace is plain files you can download through the file browser or terminal anytime&rdquo;.
OneClickClaw: &ldquo;No proprietary layer, no export fee, no conversation with support. Copy it
off and go&rdquo;. Hostinger, on switching applications: &ldquo;your configurations, conversation
history, and custom setups don&rsquo;t transfer automatically&rdquo;, and &ldquo;We recommend exporting or
noting down any important settings before switching&rdquo;. xCloud: &ldquo;No lock-in, cancel
anytime&rdquo;, and nothing about how the data leaves.</p>
<p>On Aplite the files you wrote and the files the agents wrote are plain markdown: notes,
<code>AGENT.md</code> instructions, tasks and schedules as <code>.md</code> with frontmatter, and any of them
downloads as a file from its own menu. What does not download is the runtime state:
session transcripts, the memory database, pairing records. There is no <code>openclaw backup
create</code> for you to run and no one-click archive. Moving to your own server is a fresh
install from the docs above with your files carried over, which is described in
<a href="https://aplite.ai/blog/managed-openclaw-vs-self-hosting#moving-between-the-columns">Moving between the columns</a>. Cancelling keeps the workspace
intact until the end of the billing period.</p>
<h2 id="security-gateway-and-approvals">Security: gateway and approvals</h2>
<p>Self-hosted, security is a checklist you own. The <a href="https://docs.openclaw.ai/gateway/security">security page</a>
notes that &ldquo;on a regular host install the Gateway binds to loopback&rdquo;, covers exposing it
(firewall, Tailscale, a reverse proxy), pairing codes for unknown DM senders, and
<code>openclaw security audit</code> for drift. Two defaults matter for the comparison. The
<a href="https://docs.openclaw.ai/gateway/security/trust-model">trust model</a> states that host
exec &ldquo;is allowed without approval prompts&rdquo; by default (<code>security="full"</code>, <code>ask="off"</code>),
&ldquo;intentional UX, not a vulnerability by itself&rdquo;, and that exec approvals &ldquo;are guardrails
for operator intent, not hostile multi-tenant isolation&rdquo;; for strong boundaries it says
&ldquo;Use sandboxing and host isolation&rdquo;. And <a href="https://docs.openclaw.ai/gateway/sandboxing">sandboxing</a>
&ldquo;is off by default&rdquo;, runs tool execution in Docker when on, and &ldquo;is not a perfect security
boundary, but it materially limits filesystem and process access when the model does
something dumb&rdquo;. On your own machine, add that the agent runs as you, on the disk with
your documents on it.</p>
<p>Managed hosts describe the machine, rarely the policy. OneClickClaw: &ldquo;a machine nobody
else shares&rdquo;. Hostinger and xCloud list &ldquo;security&rdquo; among what they handle. The pages do
not say whether exec approvals are on, whether the agent is sandboxed or what the Gateway
is bound to.</p>
<p>Aplite&rsquo;s answer is in its code, so it can be exact. Each workspace is one container on
its own Docker network; the gateway port is published on <code>127.0.0.1</code> of our host only, so
nothing on the internet reaches a gateway directly; device pairing is done by the
platform with one signed identity, and you never see the Control UI. The agent&rsquo;s <code>exec</code>
runs inside the container, which is the sandbox. On top of the runtime sits the
<a href="https://aplite.ai/docs/approvals">approval gate</a>: file writes, edits, shell commands that change files
and anything that sends something out are held as proposals until you accept, edit or
reject them. The trade is the one in the openclaw.json row: these policies are set for
you, and you cannot loosen them.</p>
<h2 id="always-on">Always on</h2>
<p>A VPS is on for the rent. Your own machine is on while it is awake and online; the
upstream <a href="https://docs.openclaw.ai/help/faq-first-run/providers-and-hosting">FAQ</a> lists
&ldquo;sleep/network drops disconnect it, OS updates/reboots interrupt it, must stay awake&rdquo;
as the laptop&rsquo;s disadvantages. Managed paid plans are on; the free tiers are not, by
design: Agent37&rsquo;s free agent sleeps after 15 minutes idle, Aplite&rsquo;s Free workspace
sleeps after 6 hours idle and wakes on the next visit, usually under a minute, which
means a Telegram bot on Free answers only while the workspace is awake. Pro and Power
stay on. Which free plans sleep and which take a card is in
<a href="https://aplite.ai/blog/openclaw-free-plan-no-credit-card">OpenClaw hosting with a free plan and no credit card</a>.</p>
<h2 id="price">Price</h2>
<p>The numbers are in <a href="https://aplite.ai/blog/openclaw-cost-per-month">What OpenClaw really costs per month</a>;
the shape is enough here. A VPS is $5 to $12 a month (Contabo, DigitalOcean, Hostinger as
printed on 20 October 2026) plus the model tokens, about $1.50 a month on a cheap model
and $31 on a frontier one at thirty messages a day, plus whatever the heartbeat adds.
Your own machine is electricity, under $1 to $9 a month for a Mac mini, plus the same
model line, or $0 with Ollama; the other $0 paths, Oracle&rsquo;s free tier among them, are in
<a href="https://aplite.ai/blog/openclaw-hosting-for-free">OpenClaw hosting for free</a>. Managed entry plans run $0
to $29 across the eighteen hosts, with model usage included at some. Aplite is $0 on Free with 300 welcome credits
that do not expire, $20 on Pro with 2,000 credits a month, $50 on Power with 6,000, at
1 credit = 1¢ of model use priced above OpenRouter&rsquo;s list.</p>
<h2 id="your-time">Your time</h2>
<p>The row no page prints. Self-hosted, it is the install, each update decision (three
stable releases in five days this month), the backup routine, the security checklist and
the evening the Gateway stops answering. Managed, it is agents and channels, and on
Aplite also the approvals queue, which is work you take on in exchange for an agent that
cannot change a file without you.</p>
<h2 id="what-a-managed-host-keeps-for-itself">What a managed host keeps for itself</h2>
<p>Across the pages, the managed trade has a shape. The host decides the version. The host
decides whether you get a shell, and most pages do not say. The host decides what model
accounts are possible, and the &ldquo;your API key&rdquo; column of the
<a href="https://aplite.ai/blog/managed-openclaw-hosting-compared">comparison</a> is where that lands. Backups exist
but their frequency, encryption and restore path are stated by about half the hosts. The
exit is described by four (Everpod, Agent37, OneClickClaw, Hostinger) and silent at the
rest. None of the pages read for this article says whether exec approvals are on or the
agent is sandboxed. That silence is a finding: for every row where a managed page says
nothing, the self-hosted column has a documented answer and a command.</p>
<h2 id="what-aplite-keeps-for-itself-exactly">What Aplite keeps for itself, exactly</h2>
<ul>
<li><strong>No shell for you.</strong> The agent runs commands in its container; you see proposals and
  results.</li>
<li><strong>No raw <code>openclaw.json</code>.</strong> Agents, models, skills and channels are set in the browser;
  settings the interface does not show, among them the heartbeat interval, the tool
  allowlist and the sandbox policy, are fixed by us.</li>
<li><strong>Models from the catalog only.</strong> 21 models through one metered account; no own key,
  no local models.</li>
<li><strong>Skills from the catalog, plus your own.</strong> 24 curated skills, skills your agents write,
  MCP servers you host; packages inside the container on Pro and Power only; no arbitrary
  plugins.</li>
<li><strong>Our release schedule.</strong> 2026.10.10 since 3 November 2026, the upstream stable of that
  day; on 20 October it was 2026.9.8 against upstream&rsquo;s 2026.10.5.</li>
<li><strong>No backup or restore button.</strong> Nightly encrypted backups exist; using one goes
  through us.</li>
<li><strong>No archive export.</strong> Notes and agent files download as <code>.md</code>; the runtime state does
  not.</li>
<li><strong>Free sleeps and is capped.</strong> After 6 hours idle, until the next visit; 25 credits a
  day until a pack is bought.</li>
</ul>
<p>What stays: a standard OpenClaw runtime per workspace, so what the OpenClaw docs say about
agents, sessions, skills and tools applies; files in plain markdown; your phone — Telegram in two
minutes with a BotFather token, WhatsApp and WeChat by QR; and the approvals layer, which is the one thing in this
list that self-hosting does not give you by default. Aplite is run by PIKsolution OÜ, an
Estonian company, on a server in Nuremberg, and is not affiliated with the OpenClaw
project.</p>
<h2 id="moving-between-the-columns">Moving between the columns</h2>
<p><strong>From your server to a managed host.</strong> The workspace files travel: <code>AGENTS.md</code>,
<code>SOUL.md</code>, <code>USER.md</code>, <code>MEMORY.md</code> and your notes are plain text and paste into any host
that shows you the files. The config does not, because the host writes its own; sessions
and pairing do not either. Channels are re-created on the host&rsquo;s side (a Telegram bot
token from @BotFather works anywhere).</p>
<p><strong>From a managed host to your server.</strong> It depends on what the host lets you take. With
SSH (OneClickClaw, Agent37&rsquo;s terminal) the self-hosted procedure applies on their
machine: <code>openclaw backup create</code>, copy the archive, <code>openclaw backup restore</code> on yours.
Everpod sends a zip. From Aplite you take the files: download the notes and the agents&rsquo;
instruction files, install OpenClaw on the new machine from the
<a href="https://docs.openclaw.ai/install">install page</a>, paste the instructions into the new
workspace, add your own model key, re-create the Telegram bot and approve the pairing.
Session history and the memory database stay behind, and the approvals layer does not
exist on a stock runtime; its nearest equivalents are the exec approvals and sandboxing
settings in the security section above. Nothing on Aplite&rsquo;s side blocks or charges for
leaving; the workspace stays until the billing period ends.</p>
<h2 id="which-column-by-what-you-need">Which column, by what you need</h2>
<p>The table decides itself once the requirement is specific.</p>
<ul>
<li><strong>Root, a local model, or a release on the day it ships</strong> → self-hosting; the docs
  cover the providers, and your own machine is the $0 version.</li>
<li><strong>An OpenClaw plugin the runtime does not bundle</strong> → self-hosting, or a managed host
  with a shell (OneClickClaw, Agent37); a plain npm or pip package also installs on
  Aplite&rsquo;s Pro and Power.</li>
<li><strong>A gateway that answers at three in the morning</strong> → a VPS, or a paid managed plan; not
  a laptop and not a free tier that sleeps.</li>
<li><strong>Model billing without an API key</strong> → a managed host that runs the model account
  (Aplite, Runware, Hostinger, Everpod, RapidClaw, ClawHosters, OpenClaw Launch, Ampere);
  self-hosting needs a key, a subscription login or a local model.</li>
<li><strong>Data that must not leave a machine you control</strong> → your own machine with a local
  model; every other column sends the model call to a provider.</li>
<li><strong>An exit you can rehearse</strong> → self-hosting (backup, restore), or a host whose page says
  how the data leaves (Everpod, Agent37, OneClickClaw); on Aplite the files download and
  the runtime state does not.</li>
<li><strong>A review step before the agent changes anything</strong> → Aplite&rsquo;s approvals, or
  self-hosting with exec approvals and sandboxing switched on by you.</li>
<li><strong>No server, no key, no shell, tonight</strong> → a managed host, with the list of what it
  keeps read first; Aplite&rsquo;s is above.</li>
</ul>
<p>Everything on this page was read on 20 October 2026; the vendor rows will be re-read with
the comparison once a quarter.</p>
<h2 id="faq">FAQ</h2>
<h3 id="managed-openclaw-vs-self-hosted-which-is-better">Managed OpenClaw vs self-hosted: which is better?</h3>
<p>Neither in general; the table has twelve rows and they point different ways. Self-hosting
gives you root, <code>~/.openclaw/openclaw.json</code>, any provider or a local model, updates on
your schedule and <code>openclaw backup create</code> for the exit, and asks for the install, the
updates, the backups and the security checklist in return. A managed host removes the
server and sometimes the API key, and keeps for itself some of the shell, the config
file, the model choice, the release timing and the backup controls, differently at each
host. The decision is which rows you need; the list above pairs each requirement with a
column.</p>
<h3 id="do-i-lose-anything-with-managed-hosting">Do I lose anything with managed hosting?</h3>
<p>Yes, and what you lose depends on the host, which is why the pages should be read rather
than the category. Common to most: the version is the host&rsquo;s choice and arrives after
their testing; the model account is theirs, or restricted to the providers they list;
backups happen on their schedule with their restore path. Variable: a shell (full SSH at
OneClickClaw, a web terminal at Agent37, none stated at several), access to
<code>openclaw.json</code>, plugin installation, an export route. On Aplite the list is exact and
printed above: no shell, no raw config, catalog models only, curated skills plus your
own, our release schedule, no backup or restore button, no archive export, and a Free
plan that sleeps after 6 hours idle. What you keep on any managed instance is a standard
OpenClaw runtime, and on Aplite the approvals layer on top of it.</p>
<h3 id="can-i-move-from-managed-to-my-own-server">Can I move from managed to my own server?</h3>
<p>If the host gives you a shell, yes, with the runtime&rsquo;s own tools: <code>openclaw backup create</code>
on their machine, copy the archive, <code>openclaw backup restore --target</code> on yours, then
<code>openclaw doctor</code>; the docs&rsquo; migrating guide covers the details, including re-installing
plugin dependencies and relinking WhatsApp. Hosts without a shell vary: Everpod packs a
zip on request, Agent37 lets you download the files, Hostinger says settings &ldquo;don&rsquo;t
transfer automatically&rdquo;. From Aplite you take the files and not the state: notes and
agent instruction files download as markdown, you install OpenClaw on the new machine,
paste the instructions, add your own key and re-create the Telegram bot; session history
and the memory database stay behind.</p>
<h3 id="can-i-edit-openclawjson-on-a-managed-host">Can I edit openclaw.json on a managed host?</h3>
<p>Only where the host gives you a shell on the machine: OneClickClaw includes SSH with admin
rights on every tier, Agent37 a web terminal and file browser. The other pages read for
this article do not mention the file. On Aplite you cannot: the platform writes
<code>openclaw.json</code> from the settings you make in the browser, and a setting the interface
does not expose, such as the heartbeat interval or the tool allowlist, is fixed.</p>
<h3 id="what-does-self-hosting-openclaw-involve-day-to-day">What does self-hosting OpenClaw involve day to day?</h3>
<p>The install once (Node 24.16 or newer, the one-line installer, <code>openclaw gateway install</code>
for a service), then: running <code>openclaw update</code> when a release you want appears, with a
verified backup first as the docs advise; a backup routine, by hand or with
<code>openclaw backup enable</code>, stored encrypted off the machine; the security checklist on the
gateway page and <code>openclaw security audit</code> for drift; and keeping the machine awake and
online, which on a laptop the upstream FAQ lists as the main disadvantage. The commands
are short; the recurring part is deciding, and noticing when the Gateway has stopped.</p>]]></content:encoded>
  </item>
  <item>
    <title>OpenClaw hosting with a free plan and no credit card: who offers what</title>
    <link>https://aplite.ai/blog/openclaw-free-plan-no-credit-card</link>
    <guid isPermaLink="true">https://aplite.ai/blog/openclaw-free-plan-no-credit-card</guid>
    <pubDate>Mon, 19 Oct 2026 00:00:00 +0000</pubDate>
    <description>Which OpenClaw hosts have a free plan that stays free, which have a trial, who asks for a card, who needs your API key, and when each one ends. Read on each vendor&#x27;s page, October 2026.</description>
    <content:encoded><![CDATA[<figure class="fig">
<a class="zoom" href="https://aplite.ai/assets/blog/openclaw-free-plan-no-credit-card.webp?v=8fcf5308" target="_blank" rel="noopener"><img src="https://aplite.ai/assets/blog/openclaw-free-plan-no-credit-card.webp?v=8fcf5308" alt="Fifteen hosts and tiers, sorted by the two questions that matter: is there a card at sign-up, and when does free end." width="1560" height="820" decoding="async"></a>
<figcaption>Fifteen hosts and tiers, sorted by the two questions that matter: is there a card at sign-up, and when does free end.</figcaption>
</figure>
<p>&ldquo;Free&rdquo; on an OpenClaw hosting page means one of four things: a plan that stays free, a
trial that ends in days, a cloud free tier where you install everything yourself, or free
hosting with tokens billed from the first reply. &ldquo;No credit card&rdquo; is a separate question,
and the two do not line up: some hosts take a card for a plan that never charges it, and
some trials need nothing but a Google account and then delete the server on day seven.</p>
<p>The table below answers both questions per host, plus the third one that decides whether
$0 is really $0: whether you need your own API key. Every line was read on the vendor&rsquo;s
own page on 19 October 2026, and the links go there.</p>
<h2 id="the-table">The table</h2>
<div class="tscroll"><table>
<thead>
<tr>
<th>Host</th>
<th>What is free</th>
<th>Card at sign-up</th>
<th>Your API key</th>
<th>Free ends</th>
<th>Watch for</th>
</tr>
</thead>
<tbody>
<tr>
<td><a href="https://aplite.ai/docs/openclaw">Aplite Free</a></td>
<td>a plan: one agent, the full workspace, budget models, 300 welcome credits</td>
<td>no</td>
<td>not needed, and not supported</td>
<td>never</td>
<td>the workspace sleeps after 6 hours idle and wakes on the next visit, usually under a minute; 25 credits a day until a pack is bought; no schedules; Pro is $20/mo</td>
</tr>
<tr>
<td><a href="https://www.agent37.com/personal">Agent37 Free</a></td>
<td>a plan: one agent, 1 vCPU, 4 GB</td>
<td>no</td>
<td>after &ldquo;$0.20 of the default model to start&rdquo;, yes</td>
<td>never</td>
<td>sleeps after 15 minutes idle</td>
</tr>
<tr>
<td><a href="https://runware.ai/free-openclaw">Runware</a></td>
<td>hosting; inference &ldquo;billed per token&rdquo;</td>
<td>no, &ldquo;no payment required to deploy&rdquo;</td>
<td>not needed</td>
<td>hosting never; tokens paid from the first reply, $2 credit with a business email</td>
<td>one instance per account; <a href="https://dev.to/runware/we-hosted-openclaw-so-you-dont-have-to-3698">suspended after 5 days unused</a>, data kept</td>
</tr>
<tr>
<td><a href="https://www.ampere.sh/">Ampere</a></td>
<td>FAQ: hosting during beta, &ldquo;$10 in AI credits included&rdquo;</td>
<td>FAQ: no</td>
<td>not needed on the credits; BYOK is listed on Pro and Ultra</td>
<td>after the beta</td>
<td>the plan table says &ldquo;Free&rdquo; is $1/mo with 5,000 credits, &ldquo;Start for $1&rdquo;; the page disagrees with itself</td>
</tr>
<tr>
<td><a href="https://oneclickclaw.io/openclaw">OneClickClaw</a></td>
<td>a 7-day trial on a Starter server, 2 vCPU, 4 GB, EU</td>
<td>no, a Google account</td>
<td>yes, or a ChatGPT / SuperGrok subscription sign-in</td>
<td>day 7: &ldquo;the server and its data are removed&rdquo; unless a payment method is added</td>
<td>then €14.99/mo, €12.49 on annual</td>
</tr>
<tr>
<td><a href="https://runonflux.com/apps/openclaw/pricing">Flux</a></td>
<td>the first week, new users only</td>
<td>no</td>
<td>yes</td>
<td>after 7 days, &ldquo;ends on its own&rdquo;</td>
<td>then $5.49/mo, 4 GB</td>
</tr>
<tr>
<td><a href="https://rapidclaw.dev/">RapidClaw</a></td>
<td>5 messages</td>
<td>no</td>
<td>not needed</td>
<td>after 5 messages</td>
<td>then $29/mo with $20 in API credits</td>
</tr>
<tr>
<td>KiloClaw</td>
<td>discontinued</td>
<td>—</td>
<td>—</td>
<td>—</td>
<td>end-of-life notice in Kilo&rsquo;s docs by <a href="https://web.archive.org/web/20260825173641/https://kilo.ai/docs/kiloclaw/overview">25 August 2026</a>, pages removed on <a href="https://github.com/Kilo-Org/kilocode/pull/14658">29 September 2026</a> as &ldquo;sunset&rdquo;; replacements by column in <a href="https://aplite.ai/blog/managed-openclaw-hosting-compared#alternatives-to-kiloclaw">Managed OpenClaw hosting compared</a></td>
</tr>
<tr>
<td><a href="https://docs.oracle.com/en-us/iaas/Content/FreeTier/freetier.htm">Oracle Cloud Always Free</a></td>
<td>a VM: 2 Arm OCPUs and 12 GB, or two 1 GB micro VMs, 200 GB disk</td>
<td>yes, plus a phone number; &ldquo;will not be charged unless you upgrade&rdquo;</td>
<td>yes, or a local model</td>
<td>never</td>
<td>you install everything; &ldquo;Out of host capacity&rdquo; on creation; instances idle for 7 days <a href="https://docs.oracle.com/en-us/iaas/Content/FreeTier/freetier_topic-Always_Free_Resources.htm">may be reclaimed</a></td>
</tr>
<tr>
<td><a href="https://docs.cloud.google.com/free/docs/free-cloud-features">Google Cloud free tier</a></td>
<td>one e2-micro, 1 GB, in three US regions, 30 GB disk</td>
<td>yes, a payment method</td>
<td>yes</td>
<td>never; the $300 trial credit after 90 days</td>
<td>1 GB is OpenClaw&rsquo;s stated minimum, not its recommendation</td>
</tr>
<tr>
<td><a href="https://aws.amazon.com/free/">AWS Free Tier</a></td>
<td>$100 in credits at sign-up, up to $200</td>
<td>the page does not say</td>
<td>yes</td>
<td>&ldquo;6 months after you open it or when your credits run out, whichever comes first&rdquo;</td>
<td>the account closes, not just the credit</td>
</tr>
<tr>
<td>Your own machine</td>
<td>everything</td>
<td>no</td>
<td>no, with a local model</td>
<td>never</td>
<td>sleep, updates and network drops stop the gateway</td>
</tr>
<tr>
<td><a href="https://www.hostinger.com/vps/openclaw-hosting">Hostinger</a></td>
<td>nothing; 30-day money-back</td>
<td>paid from day one</td>
<td>not needed: AI credits included, or a ChatGPT Plus / Pro subscription</td>
<td>—</td>
<td>$4.99/mo on a 24-month term, $10.99 monthly, renews at $11.99</td>
</tr>
<tr>
<td><a href="https://xcloud.host/openclaw-hosting/">xCloud</a></td>
<td>nothing</td>
<td>paid from day one</td>
<td>yes</td>
<td>—</td>
<td>$9.99 the first month, then $19.99</td>
</tr>
<tr>
<td><a href="https://everpod.ai/">Everpod</a></td>
<td>nothing</td>
<td>paid from day one</td>
<td>not needed: &ldquo;up to $10 of model usage a month&rdquo; included</td>
<td>—</td>
<td>$29/mo</td>
</tr>
</tbody>
</table></div>
<h2 id="free-plans-that-stay-free-no-card">Free plans that stay free, no card</h2>
<p>Three options cost nothing on day one and nothing on day 400.</p>
<p><strong>Aplite Free</strong> is a working workspace rather than a trial: a standard OpenClaw instance in
its own container, one agent, the markdown workspace, approvals, the agent on your phone
(Telegram from a bot token, WhatsApp and WeChat by QR), and 300 welcome credits that never expire, spent only when the agent calls a model. Spend
is capped at 25 credits a day; a credit pack lifts the cap. It is not always-on. A Free workspace stops after 6 hours without activity and starts again when
you open it, which takes under a minute, and a Telegram bot answers only while the workspace
is awake. Scheduled routines and the full model catalog are on
<a href="https://aplite.ai/docs/plans">Pro, $20/mo</a>; credit packs from $5 work on Free too.</p>
<p><strong>Agent37 Free</strong> is the closest comparison: one agent on 1 vCPU and 4 GB, no card, and it
sleeps after 15 minutes idle. The difference is the model: Agent37 gives &ldquo;$0.20 of the
default model to start&rdquo; and then expects your own keys or a paid plan, Aplite keeps
metering from its own account.</p>
<p><strong>Your own machine</strong> is free in the fullest sense and is the subject of
<a href="https://aplite.ai/blog/openclaw-hosting-for-free">OpenClaw hosting for free</a>: the install, a local model
through Ollama, and the <code>openclaw gateway install</code> that keeps it running after logout.</p>
<h2 id="free-hosting-paid-tokens">Free hosting, paid tokens</h2>
<p><strong>Runware</strong> charges nothing for the instance. Compute, storage and networking are &ldquo;included
at zero charge&rdquo;, there is &ldquo;no payment required to deploy&rdquo;, and a business email starts the
wallet at $2. Every reply then draws on that wallet at per-token rates, one instance per
account during the beta, and an instance unused for 5 days is suspended until you redeploy
it. Honest and cheap for a bot that talks a few times a day; the wallet is the plan.</p>
<p><strong>Ampere</strong> is harder to read. The FAQ says &ldquo;Free hosting during beta, $10 in AI credits
included&rdquo; and &ldquo;No credit card needed&rdquo;. The plan table on the same page lists the entry tier
as &ldquo;Free&rdquo;, priced at $1/mo, with 5,000 credits, where &ldquo;$1 = 1,000 credits&rdquo;, and the button
reads &ldquo;Start for $1&rdquo;. Take whichever the sign-up form actually shows you.</p>
<h2 id="trials-no-card-about-a-week">Trials: no card, about a week</h2>
<p><strong>OneClickClaw</strong> provisions a real Starter server in Denmark for 7 days against a Google
account. You bring a key or sign in with a ChatGPT or SuperGrok subscription. The page is
explicit about the end: &ldquo;At the end of day 7 the server and its data are removed&rdquo;, unless
you add a payment method; then €14.99/mo.</p>
<p><strong>Flux</strong> gives new users the first week of its $5.49/mo plan, no card, and the week &ldquo;ends on
its own after 7 days unless you choose to keep the server&rdquo;. Models are your own key.</p>
<p><strong>RapidClaw</strong> counts messages instead of days: 5, then $29/mo. The plan includes $20 in API
credits, so no key either side of the line.</p>
<p><strong>KiloClaw</strong> appears in most &ldquo;free trial&rdquo; lists, but Kilo discontinued it in 2026. Its docs
carried the notice &ldquo;KiloClaw is no longer available to new users, and support for the
product is ending&rdquo; by
<a href="https://web.archive.org/web/20260825173641/https://kilo.ai/docs/kiloclaw/overview">25 August</a>,
and on <a href="https://github.com/Kilo-Org/kilocode/pull/14658">29 September</a> the pages were
removed as &ldquo;sunset&rdquo;. There is no trial left to check; the hosts that sell the same thing,
a managed instance with the model account included, are listed by column in
<a href="https://aplite.ai/blog/managed-openclaw-hosting-compared#alternatives-to-kiloclaw">Managed OpenClaw hosting compared</a>.</p>
<h2 id="cloud-free-tiers-a-card-at-sign-up-you-install">Cloud free tiers: a card at sign-up, you install</h2>
<p>These are free machines, not free OpenClaw: you create the VM, open the firewall, run the
installer and bring a key or a local model, as in the
<a href="https://docs.openclaw.ai/install/oracle">upstream Oracle guide</a>.</p>
<ul>
<li><strong>Oracle Cloud Always Free</strong> is the most machine for $0: 2 Arm OCPUs and 12 GB of memory
  (older guides still quote 4 and 24), or two 1 GB micro instances, and 200 GB of block
  storage. Sign-up takes a card and a phone number; &ldquo;Your credit card will not be charged
  unless you upgrade your account.&rdquo; Two things end the fun: &ldquo;Out of host capacity&rdquo; when
  creating the Arm shape, and reclamation of instances that sit idle for 7 days.</li>
<li><strong>Google Cloud</strong> gives one e2-micro in us-west1, us-central1 or us-east1 with 30 GB of
  disk, behind a payment method. 1 GB of memory is the floor the OpenClaw FAQ names, and the
  2026.9.8 gateway sits at about 740 MB resident a minute after start. The $300 welcome
  credit lasts 90 days.</li>
<li><strong>AWS</strong> replaced its always-free micro instance with credits: $100 on sign-up, up to $200,
  and &ldquo;The account closes on its own 6 months after you open it or when your credits run
  out, whichever comes first.&rdquo;</li>
</ul>
<h2 id="no-free-plan">No free plan</h2>
<p><strong>Hostinger</strong> sells OpenClaw from $4.99/mo on a 24-month term ($10.99 month to month,
renewing at $11.99) with AI credits in the subscription or your ChatGPT Plus or Pro plan, and
a 30-day money-back guarantee. <strong>xCloud</strong> is $9.99 the first month and $19.99 after, your
own key. <strong>Everpod</strong> is $29/mo with up to $10 of model usage included. None has a free tier
or a no-card trial; they are the paid baseline the free offers above are measured against.</p>
<h2 id="hosted-and-free-three-steps">Hosted and free: three steps</h2>
<p>The Aplite path, for completeness:</p>
<ol>
<li><strong>Create an account.</strong> Email or Google, no card.</li>
<li><strong>Deploy.</strong> Pick the agent, press Deploy; under a minute later the workspace opens
   with its own OpenClaw instance, currently 2026.10.10.</li>
<li><strong>Ask.</strong> No key to paste: models run through one metered account and the welcome credits
   cover the first conversations.</li>
</ol>
<p>What differs from self-hosting, no shell, no raw <code>openclaw.json</code>, catalog models only, is
on the <a href="https://aplite.ai/docs/openclaw">OpenClaw page</a>. Aplite is an independent service and is not
affiliated with the OpenClaw project.</p>
<h2 id="faq">FAQ</h2>
<h3 id="is-there-openclaw-hosting-that-is-free-forever">Is there OpenClaw hosting that is free forever?</h3>
<p>Three kinds. Aplite Free and Agent37 Free are plans with no end date, both on a single
agent that sleeps between uses. Runware hosts for free indefinitely and bills tokens. Oracle,
Google Cloud and AWS give a free machine with a card on file, and you install OpenClaw
yourself.</p>
<h3 id="which-openclaw-hosts-do-not-ask-for-a-credit-card">Which OpenClaw hosts do not ask for a credit card?</h3>
<p>Aplite, Agent37, Runware, RapidClaw and Flux take no card to start, and OneClickClaw needs
only a Google account for its 7-day trial. Ampere&rsquo;s FAQ says no card. Oracle and Google
Cloud ask for a card they do not charge inside the free tier.</p>
<h3 id="does-the-aplite-free-plan-expire">Does the Aplite Free plan expire?</h3>
<p>No. There is no end date and the 300 welcome credits do not expire. The workspace sleeps
after 6 hours idle and wakes on your next visit; one agent, budget models, 25 credits a
day until a pack is bought, no scheduled automations.</p>
<h3 id="is-the-oracle-free-tier-really-free">Is the Oracle free tier really free?</h3>
<p>Yes, within the Always Free limits: 2 Arm OCPUs and 12 GB, or two 1 GB micro VMs. You give
a card and a phone number at sign-up and are &ldquo;not charged unless you upgrade&rdquo;. Capacity for
the Arm shape is often unavailable, and an instance idle for 7 days can be reclaimed.</p>
<h3 id="what-is-the-difference-between-a-free-plan-and-a-free-trial">What is the difference between a free plan and a free trial?</h3>
<p>A plan has no end date and usually a smaller machine, one agent or a sleep timer. A trial is
the paid product for a few days: OneClickClaw and Flux give 7, RapidClaw gives 5 messages,
and the server or its data goes away unless you pay.</p>]]></content:encoded>
  </item>
  <item>
    <title>Managed OpenClaw hosting compared (2026)</title>
    <link>https://aplite.ai/blog/managed-openclaw-hosting-compared</link>
    <guid isPermaLink="true">https://aplite.ai/blog/managed-openclaw-hosting-compared</guid>
    <pubDate>Mon, 19 Oct 2026 00:00:00 +0000</pubDate>
    <description>17 services that run OpenClaw for you, side by side: price from, free plan or trial, whether you need your own API key, what they manage, where the servers are, local models. Plus what happened to KiloClaw and what replaces it. Read on each vendor&#x27;s page, 19 October 2026.</description>
    <content:encoded><![CDATA[<figure class="fig">
<a class="zoom" href="https://aplite.ai/assets/blog/managed-openclaw-hosting-compared.webp?v=76416b16" target="_blank" rel="noopener"><img src="https://aplite.ai/assets/blog/managed-openclaw-hosting-compared.webp?v=76416b16" alt="Eighteen managed OpenClaw hosts in one table: price from, your key or theirs, where the servers are." width="1560" height="820" decoding="async"></a>
<figcaption>Eighteen managed OpenClaw hosts in one table: price from, your key or theirs, where the servers are.</figcaption>
</figure>
<p>&ldquo;Managed OpenClaw hosting&rdquo; means someone else runs the gateway: the server, the install,
the updates, usually the backups, and sometimes the model account too. Seventeen services
sell that in October 2026, and their pages answer the same six questions differently:
what the cheapest plan costs, whether anything is free, whether you need your own API
key, what exactly is managed, where the servers are, and whether a local model is an
option.</p>
<p>The table puts those six answers side by side for every host that the search engines and
AI assistants name when asked for OpenClaw hosting. Every cell was read on the vendor&rsquo;s
own page on 19 October 2026, and the host&rsquo;s name links to that page. Prices are per month
as printed, before tax; the sections under the table carry the quotes, the renewal prices
and the exceptions. The eighteenth row is KiloClaw, the host most lists still name: Kilo
sunset it in 2026, the row stays because people search for it, and the status, checked on
20 October 2026, and the replacements by column are in
<a href="https://aplite.ai/blog/managed-openclaw-hosting-compared#alternatives-to-kiloclaw">Alternatives to KiloClaw</a>. Providers that only rent a plain
VPS, and the Mac mini, are not in the table and are covered at the end.</p>
<h2 id="the-table">The table</h2>
<div class="tscroll"><table>
<thead>
<tr>
<th>Host</th>
<th>From</th>
<th>Free or trial</th>
<th>Your API key</th>
<th>What they run</th>
<th>Where</th>
<th>Local models</th>
</tr>
</thead>
<tbody>
<tr>
<td><a href="https://www.agent37.com/personal">Agent37</a></td>
<td>$0; Basic $3.99, Plus $9.99</td>
<td>Free plan: 1 agent, 1 vCPU, 4 GB, sleeps after 15 min; no card</td>
<td>yes on Free and Basic; Plus includes free models</td>
<td>the machine, a task board, web terminal, scheduled jobs</td>
<td>not stated</td>
<td>not stated</td>
</tr>
<tr>
<td><a href="https://aplite.ai/docs/openclaw">Aplite</a></td>
<td>$0; Pro $20, Power $50</td>
<td>Free plan: 1 agent, 300 credits, no card; 25 credits a day; sleeps after 6 h idle</td>
<td>no, and not supported; models billed in credits</td>
<td>server, config, pairing, upgrades; a workspace with approvals on top</td>
<td>Germany; EU company</td>
<td>no</td>
</tr>
<tr>
<td><a href="https://www.ampere.sh/">Ampere</a></td>
<td>&ldquo;Free&rdquo; at $1; Pro $39, Ultra $79</td>
<td>FAQ: free in beta, $10 in credits, no card; the plan table says $1</td>
<td>credits, or your own key at zero markup</td>
<td>&ldquo;No servers. No setup. No maintenance.&rdquo;; backup, auto update</td>
<td>not stated</td>
<td>not stated</td>
</tr>
<tr>
<td><a href="https://www.betterclaw.io/pricing">BetterClaw</a></td>
<td>$19 Basic, 1 agent, 2 vCPU, 4 GB; Pro $49</td>
<td>a Free plan in the FAQ, limits not listed; 7-day money-back</td>
<td>yes on Basic; managed keys from Pro</td>
<td>&ldquo;managed infrastructure&rdquo;; credits are minutes of uptime</td>
<td>not stated</td>
<td>Ollama named, not described</td>
</tr>
<tr>
<td><a href="https://clawhosters.com/pricing">ClawHosters</a></td>
<td>€19 Budget, 2 vCPU, 4 GB; €35, €59</td>
<td>48-hour trial, card on file</td>
<td>no: Kimi K2.5, Gemini Flash or DeepSeek included; own key optional</td>
<td>setup, auto-updates, monitoring; backups an add-on</td>
<td>Germany, &ldquo;full GDPR compliance&rdquo;</td>
<td>Ollama on your own hardware over a VPN</td>
</tr>
<tr>
<td><a href="https://clawtrust.ai/">ClawTrust</a></td>
<td>$29 Lite; $79, $159</td>
<td>5-day trial</td>
<td>yes on Lite (a free NVIDIA key); Pro has a $10 credit; AI caps $5–30</td>
<td>&ldquo;security, hosting, and technical setup&rdquo;</td>
<td>not stated</td>
<td>not stated</td>
</tr>
<tr>
<td><a href="https://everpod.ai/openclaw">Everpod</a></td>
<td>$29 incl. $10 of models; $59, $99</td>
<td>none</td>
<td>no: $10 of usage a month; own key at no markup</td>
<td>the pod, the install, tested backups, updates</td>
<td>Europe</td>
<td>not stated</td>
</tr>
<tr>
<td><a href="https://runonflux.com/apps/openclaw/pricing">Flux</a></td>
<td>$5.49, 2 vCores, 4 GB; Pro $9.49</td>
<td>first week free, no card; 30-day money-back</td>
<td>yes</td>
<td>a dedicated container, DDoS protection; upkeep not specified</td>
<td>nodes in 55 countries, most in Germany</td>
<td>&ldquo;point OpenClaw at a local LLM&rdquo;; no GPU</td>
</tr>
<tr>
<td><a href="https://www.hostinger.com/vps/openclaw-hosting">Hostinger</a></td>
<td>$4.99 on 24 months, $10.99 monthly; renews at $11.99</td>
<td>none; 30-day money-back</td>
<td>no: AI credits included; or ChatGPT Plus/Pro, or your key</td>
<td>setup, updates, backups, security; the VPS is yours</td>
<td>9 countries, 4 in the EU and UK</td>
<td>not stated</td>
</tr>
<tr>
<td>KiloClaw</td>
<td>discontinued: end-of-life notice in Kilo&rsquo;s docs by 25 August 2026, removed from the docs on 29 September; <a href="https://aplite.ai/blog/managed-openclaw-hosting-compared#not-in-the-table-and-why">details below</a></td>
<td>—</td>
<td>—</td>
<td>—</td>
<td>—</td>
<td>—</td>
</tr>
<tr>
<td><a href="https://myclaw.ai/pricing">MyClaw</a></td>
<td>$29 Lite, 2 vCPU, 4 GB; $59, $119</td>
<td>no free plan listed</td>
<td>tokens not included; Pro and up list $100–200 in credits</td>
<td>zero setup, auto-updates, daily backups</td>
<td>not stated</td>
<td>not stated</td>
</tr>
<tr>
<td><a href="https://oneclickclaw.io/openclaw">OneClickClaw</a></td>
<td>€14.99 Starter, 2 vCPU, 4 GB; €12.49 on annual</td>
<td>7-day trial, a Google account, no card; the server is deleted on day 7</td>
<td>yes, or a ChatGPT / SuperGrok sign-in</td>
<td>setup, monitoring, tested updates, daily and weekly backups, SSL; optional SSH</td>
<td>Denmark; DPA offered</td>
<td>Ollama on Hobbyist (€25.99) and Pro</td>
</tr>
<tr>
<td><a href="https://setupopenclaw.com/">OpenClaw Cloud</a> (setupopenclaw.com)</td>
<td>$19 Starter, 2 cores, 4 GB; $39, $79</td>
<td>none; limited beta, waitlist</td>
<td>credits sold separately, or your key</td>
<td>setup, maintenance, SSL, updates</td>
<td>Germany (Hetzner), &ldquo;full GDPR compliance&rdquo;</td>
<td>not stated</td>
</tr>
<tr>
<td><a href="https://openclawlaunch.com/pricing">OpenClaw Launch</a></td>
<td>$3 the first month, then $6 Lite, 1 vCPU, 2 GB; Pro $20</td>
<td>30 minutes, no card</td>
<td>no: $1 or $10 a month in credits plus free models; own key optional</td>
<td>Docker, SSL, monitoring, daily backups</td>
<td>not stated</td>
<td>not stated</td>
</tr>
<tr>
<td><a href="https://docs.openclaw.ai/install/railway">Railway</a></td>
<td>Hobby $5 incl. $5 of usage, then per second</td>
<td>a $5 credit for 30 days, no card</td>
<td>yes; the template takes only the gateway token</td>
<td>the container; volume, variables, updates and backups are yours</td>
<td>US ×2, Amsterdam, Singapore</td>
<td>not stated</td>
</tr>
<tr>
<td><a href="https://rapidclaw.dev/">RapidClaw</a></td>
<td>$29 incl. $20 in API credits; Builder $200</td>
<td>5 messages, no card</td>
<td>no: credits included</td>
<td>updates, patches, backups &ldquo;handled automatically&rdquo;</td>
<td>not stated</td>
<td>not stated</td>
</tr>
<tr>
<td><a href="https://runware.ai/free-openclaw">Runware</a></td>
<td>$0 hosting; tokens per use, $2 credit</td>
<td>hosting free, no payment to deploy; one instance</td>
<td>no</td>
<td>compute, storage, container, network, updates, security</td>
<td>EU and US; GDPR, ISO 27001, SOC 2</td>
<td>&ldquo;No local deployment&rdquo;</td>
</tr>
<tr>
<td><a href="https://xcloud.host/openclaw-hosting/">xCloud</a></td>
<td>$9.99 the first month, then $19.99; 4 vCPU, 6 GB</td>
<td>none</td>
<td>yes</td>
<td>server, security, updates, recovery; daily backups</td>
<td>31 locations, 8 in Europe</td>
<td>not stated</td>
</tr>
</tbody>
</table></div>
<h2 id="prices-the-headline-and-the-renewal">Prices: the headline and the renewal</h2>
<p>&ldquo;From&rdquo; is the cheapest plan as printed. Four hosts print a first-term price below the one
you pay afterwards: Hostinger $4.99 a month on a 24-month term, renewing at $11.99;
xCloud $9.99 for the first month, then $19.99; OpenClaw Launch $3, then $6; Ampere&rsquo;s
&ldquo;Free&rdquo; tier at $1 a month next to a FAQ that calls it free. Annual billing lowers three
others: OneClickClaw to €12.49, BetterClaw Pro to $39, MyClaw Lite to $199 a year.</p>
<p>The hosting price is half the bill. Unless the plan includes model usage, every reply the
agent writes is paid separately to a model provider, and the table&rsquo;s &ldquo;Your API key&rdquo; column
is where that line hides. Hosts that include some model usage in the price: Everpod
($10 a month), RapidClaw ($20 in credits), OpenClaw Launch ($1 or $10 a month), Agent37
Pro and Max ($5 and $20 a month), ClawTrust Pro ($10 a month), Hostinger (an amount the
page does not state), ClawHosters (three free models), Ampere (credits, two figures on
one page), Aplite (300 welcome credits on Free, 2,000 a month on Pro, 1 credit = 1¢). What
a month of models costs beyond that is its own question, which the OpenClaw docs answer
for self-hosting and the vendor pages mostly do not.</p>
<h2 id="free-plans-and-trials">Free plans and trials</h2>
<p>Five hosts have a plan with no end date at $0: Aplite (one agent, sleeps after 6 hours
idle), Agent37 (one agent, sleeps after 15 minutes), BetterClaw (named in the FAQ, limits
not listed), Runware (hosting free, every token paid) and, depending on which part of the
page you read, Ampere. Seven offer a trial: OneClickClaw and Flux a week, ClawTrust five
days, ClawHosters 48 hours with &ldquo;Your card won&rsquo;t be charged until the trial ends&rdquo;,
RapidClaw five messages, OpenClaw Launch 30 minutes, Railway a $5 credit for 30 days.
Hostinger, xCloud, Everpod, MyClaw and OpenClaw Cloud start paid. Which trials take a card
and what happens on the last day is the subject of
<a href="https://aplite.ai/blog/openclaw-free-plan-no-credit-card">OpenClaw hosting with a free plan and no credit card</a>.</p>
<h2 id="your-api-key-or-theirs">Your API key or theirs</h2>
<p>Three groups. Hosts that run models through their own account, so nothing is pasted:
Aplite, Runware (&ldquo;No keys to copy, no configs to edit&rdquo;), Hostinger (&ldquo;Pre-configured AI
credits&rdquo;), Everpod (&ldquo;up to $10 of model usage a month&rdquo;), RapidClaw, ClawHosters (&ldquo;Free Kimi
K2.5, Gemini Flash or Deepseek included&rdquo;), OpenClaw Launch and Ampere; on Agent37 that
starts at Plus, on BetterClaw at Pro, on ClawTrust at Pro. Hosts that take a subscription
sign-in instead of a key: OneClickClaw (ChatGPT or SuperGrok) and Hostinger (ChatGPT Plus
or Pro). Hosts where the key is yours: xCloud (&ldquo;You bring your own provider API key&rdquo;),
Flux (&ldquo;bring your own API keys&rdquo;), Railway, MyClaw (&ldquo;AI token usage isn&rsquo;t included in the
subscription&rdquo;), OpenClaw Cloud (&ldquo;AI credits sold separately or bring your own API keys&rdquo;),
BetterClaw Basic (&ldquo;BYOK only&rdquo;), Agent37 Free and Basic (&ldquo;$0.20 of the default model to
start&rdquo;), ClawTrust Lite (&ldquo;your free NVIDIA API key&rdquo;). Whether a key is avoidable at all,
with a local model, a subscription login or a free key, is in
<a href="https://aplite.ai/blog/openclaw-without-api-key">Can you use OpenClaw without an API key?</a>.</p>
<h2 id="what-managed-covers">What &ldquo;managed&rdquo; covers</h2>
<p>The word stretches from &ldquo;we run the container&rdquo; to &ldquo;we run the agent&rdquo;. At one end Railway
hosts the official template: the container runs, and the volume, the variables, the
updates and <code>openclaw backup create</code> are yours. In the middle are servers with OpenClaw
installed and kept current, sized in vCPU and GB and sold like a VPS: Hostinger (&ldquo;we
handle the setup, updates, backups, and security&rdquo;), xCloud (&ldquo;the server, security,
updates, and recovery&rdquo;), OneClickClaw (monitoring every five minutes, a snapshot before
every update, optional SSH), ClawHosters, Flux, MyClaw, OpenClaw Cloud (&ldquo;server setup,
maintenance, SSL certificates, and updates&rdquo;). At the other end the page sells the running
agent and names what is handled rather than the machine: Aplite, Runware, Agent37,
Everpod (&ldquo;restores we actually test&rdquo;), RapidClaw (&ldquo;Updates, security patches, and backups
are all handled automatically&rdquo;), OpenClaw Launch, BetterClaw, ClawTrust.</p>
<p>Backups, where the page says so: daily on xCloud, Everpod, MyClaw, OpenClaw Launch and
RapidClaw; daily and weekly on OneClickClaw; included on Hostinger; an add-on on
ClawHosters&rsquo; pricing page; not stated on Runware. What a managed instance takes away is
the other half of the column: on Aplite there is no shell, no raw <code>openclaw.json</code> and no
model outside the catalog, and the list is on the <a href="https://aplite.ai/docs/openclaw">OpenClaw page</a>; other
hosts draw the line elsewhere, and only OneClickClaw and Agent37 say on the page that a
terminal is included.</p>
<h2 id="where-the-servers-are">Where the servers are</h2>
<p>Eleven pages name a place. Germany: Aplite (Nuremberg), ClawHosters (&ldquo;All servers are
located in Germany&rdquo;), OpenClaw Cloud (&ldquo;Hosted on German servers (Hetzner)&rdquo;). Denmark:
OneClickClaw (&ldquo;A private server in Denmark that only you use&rdquo;). &ldquo;Pods are hosted in
Europe&rdquo;: Everpod. &ldquo;EU &amp; US data residency&rdquo;: Runware. Your choice at order time, with EU
options: Hostinger (France, Germany, Lithuania and the UK among nine countries), xCloud
(Frankfurt, Amsterdam, Paris, Madrid, Stockholm, Warsaw, London and Manchester among 31
locations), Railway (Amsterdam among four regions). Flux runs on independent nodes in 55
countries, most of them in Germany, and does not describe picking one. Seven say nothing:
Agent37, Ampere, BetterClaw, ClawTrust, MyClaw, OpenClaw Launch, RapidClaw.</p>
<p>GDPR appears on three pages: ClawHosters, OpenClaw Cloud and Runware. OneClickClaw offers
a data processing agreement, BetterClaw links a DPA and a sub-processor list, Aplite&rsquo;s
<a href="https://aplite.ai/privacy">privacy policy</a> lists every processor. One thing an EU server does not change:
the model call leaves it. On Aplite, model requests go to the provider you picked through
OpenRouter in the US, and the same is true of any host that bills frontier models.</p>
<h2 id="local-models">Local models</h2>
<p>Four pages mention a model that is not a cloud API. OneClickClaw runs Ollama on its
Hobbyist and Pro servers, Llama 3.2 1B and 3B. ClawHosters connects Ollama on your own
hardware through a ZeroTier VPN. Flux says you can point OpenClaw at a local LLM, on
containers of 4 or 8 GB with no GPU. BetterClaw lists &ldquo;local via Ollama&rdquo; among model
options without saying where it runs. Two say no: Aplite (catalog models only) and Runware
(&ldquo;No local deployment&rdquo;). The other twelve do not mention it. The $0-token path, Ollama on
a machine you own, is in <a href="https://aplite.ai/blog/openclaw-hosting-for-free">OpenClaw hosting for free</a>.</p>
<h2 id="not-in-the-table-and-why">Not in the table, and why</h2>
<ul>
<li><strong>KiloClaw: discontinued.</strong> Kilo&rsquo;s hosted OpenClaw was marked end of life in the
  summer of 2026 and removed from Kilo&rsquo;s documentation on 29 September 2026. The docs
  page carried the notice &ldquo;KiloClaw is end of life. KiloClaw is no longer available to new
  users, and support for the product is ending&rdquo; by
  <a href="https://web.archive.org/web/20260825173641/https://kilo.ai/docs/kiloclaw/overview">25 August 2026</a>
  (the 23 July snapshot has no such notice), and the pull request that deleted the pages,
  <a href="https://github.com/Kilo-Org/kilocode/pull/14658">merged 29 September</a>, says &ldquo;Both
  products have been sunset&rdquo;. There is no post about it on the Kilo blog; <code>kilo.ai/kiloclaw</code>
  has redirected to the Kilo Code homepage since late August; on 20 October 2026 the Kilo
  status page still listed KiloClaw as operational. A date for switching existing
  instances off, 20 October 2026, appears on one user&rsquo;s blog and in Reddit reports of an
  email; Kilo has not published one.</li>
<li><strong>ClawTrust</strong> is at clawtrust.ai; clawtrust.com, which some lists link, is a parked
  domain.</li>
<li><strong>&ldquo;OpenClaw Cloud&rdquo;</strong> is the brand of setupopenclaw.com, a third party. The project&rsquo;s own
  site says &ldquo;No subscription. No hosted tier. No token.&rdquo; and &ldquo;There is no enterprise
  edition and no paid version&rdquo;; the official <a href="https://docs.openclaw.ai/vps">install guides</a>
  cover hosting on third-party providers, not a service of its own.</li>
<li><strong>Hetzner, DigitalOcean, Oracle Cloud, Contabo and the other VPS providers</strong> rent a
  machine; you install and run OpenClaw on it. They are the self-hosting baseline the
  table&rsquo;s prices sit next to, and the upstream docs have an install page for several.
  The $0 version of that path, including Oracle&rsquo;s Always Free tier, is in
  <a href="https://aplite.ai/blog/openclaw-hosting-for-free">OpenClaw hosting for free</a>.</li>
<li><strong>A Mac mini</strong> is the same baseline on your desk: no monthly fee, and sleep, updates and
  network drops are yours to handle.</li>
</ul>
<h2 id="alternatives-to-kiloclaw">Alternatives to KiloClaw</h2>
<p>KiloClaw&rsquo;s own documentation described it as a one-click OpenClaw instance where the &ldquo;API
key is automatically generated and refreshed&rdquo;, the agent &ldquo;runs on your Kilo Gateway
balance&rdquo; and several free models are included: hosted, no key to paste, models billed
through the host. The table&rsquo;s columns map that onto what is still sold. No key to paste,
because the host runs the model account: Aplite, Runware, Hostinger, Everpod, RapidClaw,
ClawHosters, OpenClaw Launch and Ampere, plus Agent37 from Plus, BetterClaw from Pro and
ClawTrust from Pro. A plan at $0 with no end date: Aplite, Agent37, BetterClaw and
Runware; a trial instead: OneClickClaw and Flux for a week, ClawTrust five days,
ClawHosters 48 hours, RapidClaw five messages, OpenClaw Launch 30 minutes, Railway a $5
credit. Servers in a named EU country: Aplite, ClawHosters and OpenClaw Cloud (Germany),
OneClickClaw (Denmark), Everpod (Europe), Runware (EU and US); Hostinger, xCloud and
Railway let you pick the region at order time. Entry prices run from $0 to $29 a month,
and the &ldquo;From&rdquo; column has each one.</p>
<h2 id="the-aplite-row-in-full">The Aplite row, in full</h2>
<p>For the reader who came for the comparison and wants the one line expanded: Aplite runs a
standard OpenClaw instance per workspace in its own container on a server in Nuremberg,
with models from a catalog over one metered account, billed in credits, no key to paste
and no option to bring one. Free is one agent, budget models and 300 welcome credits that
do not expire, no card; spend is capped at 25 credits a day until a pack is bought; the
workspace sleeps after 6 hours idle and wakes on the next visit, usually under a minute. Pro is $20 a month with 2,000 credits, six agents, always-on
and scheduled routines; Power is $50. Around the runtime sits a markdown workspace,
approvals for every file change, schedules and triggers, published pages, and the agent on
your phone through Telegram, WhatsApp or WeChat. Not included: a shell, the raw config, local models, and releases on the day they ship. The
full list, both directions, is on the <a href="https://aplite.ai/docs/openclaw">OpenClaw page</a>. Aplite is run by
PIKsolution OÜ, an Estonian company, is independent, and is not affiliated with the
OpenClaw project.</p>
<p>Prices on this page move. The first line of the article carries the date they were read;
the plan is to re-read every row once a quarter.</p>
<h2 id="faq">FAQ</h2>
<h3 id="what-does-managed-openclaw-hosting-mean">What does &ldquo;managed OpenClaw hosting&rdquo; mean?</h3>
<p>A service runs the OpenClaw gateway for you: the server, the install, the updates and
usually the backups, so you configure agents and channels instead of a Linux box. Some
hosts also run the model account, so there is no API key to paste. The opposite is
self-hosting: a VPS from Hetzner or DigitalOcean, or a Mac mini, with OpenClaw installed
and maintained by you.</p>
<h3 id="how-much-does-managed-openclaw-hosting-cost-per-month">How much does managed OpenClaw hosting cost per month?</h3>
<p>In October 2026, from $0 to about $30 for the entry plan: Aplite, Agent37, Runware and
BetterClaw have a free tier; Hostinger starts at $4.99 on a 24-month term and renews at
$11.99; Flux is $5.49; OpenClaw Launch $6 after the first month; xCloud $19.99 after the
first month; OneClickClaw €14.99; ClawHosters and OpenClaw Cloud €19 and $19; Everpod,
MyClaw, RapidClaw and ClawTrust $29. Model usage is extra unless the plan includes credits.</p>
<h3 id="which-managed-openclaw-hosts-do-not-need-an-api-key">Which managed OpenClaw hosts do not need an API key?</h3>
<p>Aplite, Runware, Hostinger, Everpod, RapidClaw, ClawHosters, OpenClaw Launch and Ampere
run models through their own account. Agent37 from Plus, BetterClaw from Pro and
ClawTrust from Pro include model access too. OneClickClaw and Hostinger also accept a
ChatGPT subscription sign-in. xCloud, Flux, Railway, MyClaw and OpenClaw Cloud expect your
own key.</p>
<h3 id="which-managed-openclaw-hosts-are-in-the-eu">Which managed OpenClaw hosts are in the EU?</h3>
<p>Pages that name a place: Aplite (Germany), ClawHosters (Germany), OpenClaw Cloud
(Germany, Hetzner), OneClickClaw (Denmark), Everpod (Europe), Runware (EU and US).
Hostinger, xCloud and Railway let you pick an EU location at order time. Seven hosts do
not say where their servers are. An EU server does not keep model calls in the EU; those
go to the model provider.</p>
<h3 id="is-there-an-official-openclaw-cloud">Is there an official OpenClaw cloud?</h3>
<p>No. The OpenClaw site says &ldquo;No subscription. No hosted tier. No token.&rdquo; The service sold
as &ldquo;OpenClaw Cloud&rdquo; is setupopenclaw.com, a third party in limited beta. Every host in the
table, Aplite included, is independent of the OpenClaw project.</p>
<h3 id="which-managed-openclaw-hosts-have-a-free-plan">Which managed OpenClaw hosts have a free plan?</h3>
<p>Aplite (one agent, 300 credits, sleeps after 6 hours idle), Agent37 (one agent, sleeps
after 15 minutes), BetterClaw (a Free plan named in its FAQ) and Runware (free hosting,
tokens billed per use). Ampere&rsquo;s FAQ says free during beta while its plan table says $1 a
month. Trials rather than plans: OneClickClaw and Flux for 7 days, ClawTrust 5 days,
ClawHosters 48 hours, RapidClaw 5 messages, OpenClaw Launch 30 minutes.</p>
<h3 id="what-are-the-alternatives-to-kiloclaw">What are the alternatives to KiloClaw?</h3>
<p>KiloClaw, Kilo&rsquo;s hosted OpenClaw, carried an end-of-life notice in its docs by 25 August
2026 and was removed from Kilo&rsquo;s documentation as &ldquo;sunset&rdquo; on 29 September 2026. It sold a
hosted instance with the model account included and no API key to paste. The hosts in the
table that work the same way: Aplite, Runware, Hostinger, Everpod, RapidClaw, ClawHosters,
OpenClaw Launch and Ampere, plus Agent37, BetterClaw and ClawTrust on their higher plans.
Free plans at Aplite, Agent37, BetterClaw and Runware; servers in the EU at Aplite,
ClawHosters, OpenClaw Cloud, OneClickClaw and Everpod, or by choice at Hostinger, xCloud
and Railway.</p>]]></content:encoded>
  </item>
  <item>
    <title>Can you use OpenClaw without an API key? Every path, compared</title>
    <link>https://aplite.ai/blog/openclaw-without-api-key</link>
    <guid isPermaLink="true">https://aplite.ai/blog/openclaw-without-api-key</guid>
    <pubDate>Mon, 19 Oct 2026 00:00:00 +0000</pubDate>
    <description>Yes, four ways: a local model, a ChatGPT, Claude or Copilot subscription login, a free key, or a host that meters models for you. What each costs and where it stops, checked October 2026.</description>
    <content:encoded><![CDATA[<figure class="fig">
<a class="zoom" href="https://aplite.ai/assets/blog/openclaw-without-api-key.webp?v=668cdcde" target="_blank" rel="noopener"><img src="https://aplite.ai/assets/blog/openclaw-without-api-key.webp?v=668cdcde" alt="The error a fresh install prints, and the four ways past it: a local model, a subscription login, a free key, a host that holds the key." width="1560" height="820" decoding="async"></a>
<figcaption>The error a fresh install prints, and the four ways past it: a local model, a subscription login, a free key, a host that holds the key.</figcaption>
</figure>
<p>OpenClaw itself never asks for money. Every reply it writes is a model call, though, and
the API key is where that bill goes. So &ldquo;without an API key&rdquo; is really two questions: can
the tokens be free, and can someone other than you hold the key. Both have answers. A fresh
install makes the question concrete the first time you send a message with nothing
configured:</p>
<pre><code>No API key found for provider "openrouter".
Auth store: ~/.openclaw/agents/main/agent/openclaw-agent.sqlite
Configure an API key (openclaw models auth paste-api-key --provider openrouter)
| missing-provider-auth
</code></pre>
<p>Here is every way out of that error, with what each one costs and where it stops. Commands
are from OpenClaw 2026.9.8; the upstream docs and every vendor page were read on
19 October 2026, and the links go to those pages.</p>
<h2 id="the-short-answer">The short answer</h2>
<div class="tscroll"><table>
<thead>
<tr>
<th>Path</th>
<th>Key to paste</th>
<th>Tokens cost</th>
<th>Runs on</th>
<th>The catch</th>
</tr>
</thead>
<tbody>
<tr>
<td>A local model (<a href="https://docs.openclaw.ai/providers/ollama">Ollama</a>, LM Studio, llama.cpp)</td>
<td>none</td>
<td>$0</td>
<td>your machine</td>
<td>the smallest managed llama.cpp recipe has an <a href="https://docs.openclaw.ai/gateway/local-models">8 GiB memory floor</a>; small models truncate context and raise prompt-injection risk</td>
</tr>
<tr>
<td>Your ChatGPT subscription (<a href="https://docs.openclaw.ai/providers/openai/setup">Codex OAuth</a>)</td>
<td>none, you sign in</td>
<td>your Plus or Pro plan</td>
<td>anywhere</td>
<td>one allowance shared with ChatGPT itself; eligibility is OpenAI&rsquo;s call</td>
</tr>
<tr>
<td>Your Claude subscription (<a href="https://docs.openclaw.ai/providers/anthropic">Claude CLI or setup-token</a>)</td>
<td>none, a token from Claude Code</td>
<td>your Pro or Max plan</td>
<td>the host where Claude Code is logged in, or any host with a setup-token</td>
<td>Anthropic&rsquo;s terms call API keys &ldquo;the preferred way&rdquo; for third-party tools and reserve the right to bill usage credits instead</td>
</tr>
<tr>
<td>Your <a href="https://docs.openclaw.ai/providers/github-copilot">GitHub Copilot</a> plan</td>
<td>none, device login</td>
<td>your Copilot plan</td>
<td>anywhere</td>
<td>which models you see depends on the plan and on org policy</td>
</tr>
<tr>
<td>A free key (<a href="https://openrouter.ai/docs/api-reference/limits">OpenRouter <code>:free</code></a>, <a href="https://ai.google.dev/gemini-api/docs/pricing">Gemini API free tier</a>, <a href="https://docs.openclaw.ai/cli/promos">promos</a>)</td>
<td>yes, a free one</td>
<td>$0</td>
<td>anywhere</td>
<td>50 requests a day on OpenRouter until you have bought $10; Gemini&rsquo;s free tier uses your content &ldquo;to improve our products&rdquo;; promos expire</td>
</tr>
<tr>
<td>A host that holds the key (<a href="https://aplite.ai/docs/openclaw">Aplite</a>, Runware, Hostinger, Ampere, Everpod…)</td>
<td>none</td>
<td>credits or per token, billed by the host</td>
<td>their server</td>
<td>you pick from the host&rsquo;s catalog, and a starter credit runs out</td>
</tr>
<tr>
<td>Bring your own key (self-hosting, xCloud, OneClickClaw, Flux…)</td>
<td>yes</td>
<td>the provider&rsquo;s price</td>
<td>anywhere</td>
<td>this is the path the question is trying to avoid</td>
</tr>
</tbody>
</table></div>
<h2 id="1-a-local-model-no-key-no-token-bill">1. A local model: no key, no token bill</h2>
<p>The only path where nothing is billed per token. Ollama is the easiest runtime; LM Studio,
llama.cpp and vLLM work through the same
<a href="https://docs.openclaw.ai/gateway/local-models">local models</a> page.</p>
<ol>
<li>
<p>Install Ollama from <a href="https://ollama.com/download">ollama.com/download</a> and pull a model:</p>
<pre><code>ollama pull gemma4
</code></pre>
</li>
<li>
<p>Run onboarding and pick <strong>Ollama</strong>, then <strong>Local only</strong>:</p>
<pre><code>openclaw onboard
</code></pre>
<p>Or set the model later:</p>
<pre><code>openclaw models set ollama/gemma4
</code></pre>
</li>
<li>
<p>The key field is a formality for local hosts. The
   <a href="https://docs.openclaw.ai/providers/ollama/setup">Ollama setup</a> page says loopback and
   private-network URLs &ldquo;do not need a real bearer token&rdquo; and that &ldquo;any value works&rdquo;:</p>
<pre><code>export OLLAMA_API_KEY="ollama-local"
</code></pre>
</li>
</ol>
<p>Pick Ollama with the server not running and onboarding stops with
<code>Ollama could not be reached at http://127.0.0.1:11434</code>. The limit is hardware and model
quality: the upstream FAQ answers &ldquo;Is a local model okay for casual chats?&rdquo; with &ldquo;Usually
not&rdquo;, because small or quantized models truncate context and raise prompt-injection risk.
A 1 GB cloud machine is out; a desktop with 16 GB or more is in.</p>
<h2 id="2-your-chatgpt-subscription">2. Your ChatGPT subscription</h2>
<p>OpenClaw signs in to a ChatGPT account and spends that plan&rsquo;s allowance instead of a key:</p>
<pre><code>openclaw models auth login --provider openai
</code></pre>
<p>Add <code>--device-code</code> on a headless server. OpenClaw&rsquo;s docs say OpenAI &ldquo;explicitly allows
subscription OAuth usage&rdquo; in tools like OpenClaw, and OpenAI&rsquo;s own
<a href="https://developers.openai.com/siwc/quickstart">developer page</a> says &ldquo;Eligible ChatGPT
Plus and Pro users can use their ChatGPT plan for AI requests in participating apps&rdquo; and
that &ldquo;ChatGPT plan usage is available to all open-source partners&rdquo;. The catch is that it
is one allowance: what the agent spends, ChatGPT no longer has, and the
<a href="https://docs.openclaw.ai/help/faq-first-run/providers-and-hosting">upstream FAQ</a> notes
that &ldquo;Codex OAuth uses plan-dependent quota windows that can differ from the ChatGPT
website or app&rdquo;.</p>
<h2 id="3-your-claude-subscription">3. Your Claude subscription</h2>
<p>Two token paths, both coming from Claude Code rather than from a key:</p>
<ol>
<li><strong>Claude CLI reuse.</strong> Claude Code is logged in on the same host; <code>openclaw onboard</code>
   offers it as a choice. OpenClaw prefers this path when it is available.</li>
<li>
<p><strong>Setup token.</strong> On any machine with Claude Code:</p>
<pre><code>claude setup-token
</code></pre>
<p>prints a long-lived token starting with <code>sk-ant-oat01-</code>, then:</p>
<pre><code>openclaw models auth login --provider anthropic --method setup-token
</code></pre>
</li>
</ol>
<p>The policy is the part to read before relying on it. OpenClaw&rsquo;s
<a href="https://docs.openclaw.ai/concepts/oauth">OAuth page</a> says &ldquo;Anthropic staff told us this
usage is allowed again&rdquo; and treats it as sanctioned &ldquo;unless Anthropic publishes a new
policy&rdquo;. Anthropic&rsquo;s own <a href="https://support.claude.com/en/articles/13189465-log-in-to-your-claude-account">help article</a>
says API keys are &ldquo;the preferred way to access Anthropic services using third-party
software, tools, or services&rdquo;, that Anthropic &ldquo;may at its discretion allow paid subscribers
who have enabled usage credits to use certain third-party tools&rdquo;, and that it &ldquo;reserves the
right to draw use of such third-party tools from usage credits rather than subscription
limits&rdquo;. The <a href="https://support.claude.com/en/articles/15036540-use-the-claude-agent-sdk-with-your-claude-plan">Agent SDK article</a>,
updated 7 October 2026, says &ldquo;You can still use the Claude Agent SDK, <code>claude -p</code>, and
third-party apps with your subscription limits.&rdquo; It works today for a personal agent; the
discretion is Anthropic&rsquo;s, and OpenClaw&rsquo;s advice is: &ldquo;For shared production automation,
use an Anthropic API key instead of Claude CLI.&rdquo;</p>
<h2 id="4-your-github-copilot-plan">4. Your GitHub Copilot plan</h2>
<pre><code>openclaw models auth login-github-copilot
</code></pre>
<p>runs GitHub&rsquo;s device flow; the same login also serves embeddings for memory search. The
<a href="https://docs.openclaw.ai/providers/github-copilot">Copilot page</a> says &ldquo;Copilot model
availability depends on your GitHub plan and organization policy&rdquo;, so what you can pick is
whatever GitHub&rsquo;s model picker shows your account.</p>
<p>There is no Google equivalent: the FAQ says OpenClaw &ldquo;does not offer new Gemini CLI OAuth
or Antigravity OAuth setup&rdquo;, so Gemini means an AI Studio key. The Qwen, MiniMax and Z.AI
coding plans are flat-price subscriptions that still hand you a key to paste.</p>
<h2 id="5-a-free-key-is-still-a-key">5. A free key is still a key</h2>
<p>Three ways to get one for $0, each with a ceiling:</p>
<ul>
<li><strong>OpenRouter <code>:free</code> models.</strong> Paste a key from openrouter.ai; the free variants are
  limited to <a href="https://openrouter.ai/docs/api-reference/limits">20 requests a minute and 50 a day</a>
  until you have bought $10 of credits, then 1,000 a day. One agent turn with tools is
  several requests, so 50 a day is a short afternoon.</li>
<li><strong>Gemini API free tier.</strong> Flash models are listed &ldquo;Free of charge&rdquo; on the
  <a href="https://ai.google.dev/gemini-api/docs/pricing">pricing page</a>, with the free tier marked
  &ldquo;Content used to improve our products&rdquo; and the paid tier &ldquo;not used&rdquo;.</li>
<li><strong>Promos.</strong> <code>openclaw promos list</code> shows time-limited free model offers published on
  ClawHub; <code>openclaw promos claim &lt;slug&gt;</code> sets the provider up and prints a sign-up URL for
  a free key. When the window ends &ldquo;the provider stops serving the free models&rdquo;.</li>
</ul>
<h2 id="6-a-host-that-holds-the-key">6. A host that holds the key</h2>
<p>Managed hosts that run models through their own account, so there is nothing to paste.
Prices as each page read on 19 October 2026:</p>
<ul>
<li><strong><a href="https://aplite.ai/docs/openclaw">Aplite</a>.</strong> Models from the catalog over one metered account, billed
  in credits, 1 credit = 1¢; bringing your own key is not supported. The Free plan is one
  agent, budget models, 300 welcome credits and a 25-credit daily cap, no card.</li>
<li><strong><a href="https://runware.ai/free-openclaw">Runware</a>.</strong> Hosting free, inference &ldquo;billed per
  token&rdquo;; $2 in credit with a business email, no payment to deploy.</li>
<li><strong><a href="https://www.hostinger.com/vps/openclaw-hosting">Hostinger</a>.</strong> &ldquo;Pre-configured AI
  credits&rdquo; in the subscription, amount not stated on the page, topped up through the
  <a href="https://www.hostinger.com/ai-router">AI Router</a> from $5.99 for 5 credits; or &ldquo;Use your
  ChatGPT Plus or Pro subscription&rdquo;; from $4.99/mo on a 24-month term.</li>
<li><strong><a href="https://www.ampere.sh/">Ampere</a>.</strong> The FAQ: free hosting during beta, &ldquo;$10 in AI
  credits included&rdquo;, no card; the plan table lists &ldquo;Free&rdquo; at $1/mo with 5,000 credits at
  &ldquo;$1 = 1,000 credits&rdquo;. The two disagree.</li>
<li><strong><a href="https://everpod.ai/">Everpod</a>.</strong> $29/mo covering &ldquo;up to $10 of model usage a month&rdquo;.</li>
<li><strong><a href="https://rapidclaw.dev/">RapidClaw</a>.</strong> 5 free messages, then $29/mo with $20 in API
  credits.</li>
<li><strong><a href="https://www.agent37.com/personal">Agent37</a>.</strong> Free plan with &ldquo;$0.20 of the default
  model to start&rdquo;, then your own keys or a paid plan; Plus ($9.99/mo) bundles free models.</li>
<li><strong><a href="https://oneclickclaw.io/openclaw">OneClickClaw</a>.</strong> No credits: bring a key, or sign in
  with a ChatGPT or SuperGrok subscription.</li>
<li><strong>KiloClaw.</strong> Discontinued. Kilo&rsquo;s docs carried the notice &ldquo;KiloClaw is no longer
  available to new users, and support for the product is ending&rdquo; by
  <a href="https://web.archive.org/web/20260825173641/https://kilo.ai/docs/kiloclaw/overview">25 August 2026</a>,
  and on <a href="https://github.com/Kilo-Org/kilocode/pull/14658">29 September 2026</a> the pages
  were removed as &ldquo;sunset&rdquo;. It sold exactly this path, a hosted instance with the model
  account included; the hosts that still do are listed by column in
  <a href="https://aplite.ai/blog/managed-openclaw-hosting-compared#alternatives-to-kiloclaw">Managed OpenClaw hosting compared</a>.</li>
</ul>
<p>On Aplite the whole path is three steps: create an account with email or Google, press
<strong>Deploy</strong>, ask. The usage shows as tokens under each reply and as money in Settings → Cost.
What a managed instance gives up, the shell, raw <code>openclaw.json</code> and local models, and how
the Free plan sleeps between uses, is on the <a href="https://aplite.ai/docs/openclaw">OpenClaw page</a>; the $0
self-hosted path is in <a href="https://aplite.ai/blog/openclaw-hosting-for-free">OpenClaw hosting for free</a>.
Aplite is an independent service and is not affiliated with the OpenClaw project.</p>
<h2 id="faq">FAQ</h2>
<h3 id="can-i-run-openclaw-without-any-api-key-at-all">Can I run OpenClaw without any API key at all?</h3>
<p>Yes, with a local model: install Ollama, <code>ollama pull gemma4</code>, pick Ollama in
<code>openclaw onboard</code>. Nothing is billed per token and no key is pasted; the cost is the
machine, and small models answer worse than hosted ones. Every other path either signs in
with a subscription or lets a host hold the key.</p>
<h3 id="can-i-use-openclaw-with-my-chatgpt-subscription">Can I use OpenClaw with my ChatGPT subscription?</h3>
<p>Yes. <code>openclaw models auth login --provider openai</code> signs in with ChatGPT and spends the
plan&rsquo;s allowance; OpenAI&rsquo;s developer docs say eligible Plus and Pro users can use their plan
in participating apps, and open-source clients are included. The allowance is shared with
ChatGPT itself.</p>
<h3 id="can-i-use-openclaw-with-claude-pro-or-max">Can I use OpenClaw with Claude Pro or Max?</h3>
<p>It works through Claude CLI reuse or a <code>claude setup-token</code>, and OpenClaw&rsquo;s docs say
Anthropic has told them the usage is allowed. Anthropic&rsquo;s help center calls API keys the
preferred way for third-party tools and reserves the right to bill such use from usage
credits rather than plan limits. For anything unattended, OpenClaw recommends an API key.</p>
<h3 id="why-does-openclaw-say-no-api-key-found-for-provider">Why does OpenClaw say &ldquo;No API key found for provider&rdquo;?</h3>
<p>The default model points at a provider with no auth profile. Paste a key with
<code>openclaw models auth paste-api-key --provider &lt;id&gt;</code>, sign in with
<code>openclaw models auth login --provider &lt;id&gt;</code>, or switch to a model whose provider is set
up, for example <code>openclaw models set ollama/gemma4</code>.</p>
<h3 id="which-openclaw-hosts-do-not-need-my-api-key">Which OpenClaw hosts do not need my API key?</h3>
<p>Hosts that meter models through their own account: Aplite, Runware, Hostinger, Ampere,
Everpod, RapidClaw, and Agent37 from its Plus plan up. xCloud, OneClickClaw, Flux, ClawdHost and
ClawHosters expect your own key, though OneClickClaw and Hostinger also take a ChatGPT
subscription sign-in.</p>]]></content:encoded>
  </item>
  <item>
    <title>OpenClaw hosting for free: what each option actually gives you</title>
    <link>https://aplite.ai/blog/openclaw-hosting-for-free</link>
    <guid isPermaLink="true">https://aplite.ai/blog/openclaw-hosting-for-free</guid>
    <pubDate>Wed, 14 Oct 2026 00:00:00 +0000</pubDate>
    <description>Free OpenClaw hosting has four shapes: your own machine, a free cloud tier, a managed trial, a free plan. What each covers, when it ends, and a $0 setup.</description>
    <content:encoded><![CDATA[<figure class="fig">
<a class="zoom" href="https://aplite.ai/assets/blog/openclaw-hosting-for-free.webp?v=66e511d5" target="_blank" rel="noopener"><img src="https://aplite.ai/assets/blog/openclaw-hosting-for-free.webp?v=66e511d5" alt="Two ways to the same gateway: install it yourself for $0, or a managed instance set up in under a minute." width="1560" height="820" decoding="async"></a>
<figcaption>Two ways to the same gateway: install it yourself for $0, or a managed instance set up in under a minute.</figcaption>
</figure>
<p>Search for <em>openclaw hosting free</em> and the first page is three kinds of result: hosts whose
&ldquo;free&rdquo; is a trial or a first-month price, Oracle free-tier guides (several still quoting an
Arm allowance Oracle has since halved), and &ldquo;best hosting&rdquo; lists written by the hosts. What
none of them does is the accounting. An agent that works around the clock costs two things —
a machine that stays on, and model tokens for every reply — and most free offers cover one
of the two, for a while.</p>
<p>Here is that accounting as of October 2026, and the one path that costs nothing, with every
command run on OpenClaw 2026.9.8. If you would rather not run a server at all, the same
runtime is <a href="https://aplite.ai/blog/openclaw-hosting-for-free#hosted-the-same-runtime-in-three-steps">hosted in three steps</a> further down.</p>
<h2 id="what-free-has-to-cover">What &ldquo;free&rdquo; has to cover</h2>
<p><strong>The server.</strong> The gateway is a Node process that has to stay up. The upstream
<a href="https://docs.openclaw.ai/help/faq-first-run/providers-and-hosting">hardware FAQ</a> puts the
floor at 1 vCPU, 1 GB of RAM and about 500 MB of disk, and recommends 2 GB or more. The floor
is real: capped at 1 GB, the 2026.9.8 gateway settles at about 740 MB of resident memory a
minute after start and logs <code>memory pressure: level=warning</code>. The smallest free cloud
machines give you exactly 1 GB.</p>
<p><strong>The model.</strong> Every reply is a model call, billed per token wherever the gateway runs.
Three free ways: a local model through <a href="https://docs.openclaw.ai/providers/ollama">Ollama</a>
(needs a capable machine; the same FAQ warns that small quantized models truncate context
and raise prompt-injection risk); OpenRouter&rsquo;s <code>:free</code> variants, limited to
<a href="https://openrouter.ai/docs/api-reference/limits">20 requests a minute and 50 a day</a> until
you have bought $10 of credits, then 1,000 a day; or the
<a href="https://ai.google.dev/gemini-api/docs/pricing">Gemini API free tier</a>, where Flash is free
of charge and, in Google&rsquo;s words, &ldquo;content used to improve our products&rdquo;. Hosted plans that
include credits cover tokens for as long as the credits last.</p>
<h2 id="the-options-side-by-side">The options, side by side</h2>
<p>Prices and terms as each page read on 13 October 2026; the links go to those pages.</p>
<div class="tscroll"><table>
<thead>
<tr>
<th>Option</th>
<th>Price, October 2026</th>
<th>You set up</th>
<th>Stops being free</th>
<th>Watch for</th>
</tr>
</thead>
<tbody>
<tr>
<td>Your own machine</td>
<td>$0</td>
<td>install, a model, keeping it awake</td>
<td>never</td>
<td>sleep, updates and network drops stop the gateway</td>
</tr>
<tr>
<td><a href="https://docs.oracle.com/en-us/iaas/Content/FreeTier/freetier.htm">Oracle Cloud Always Free</a></td>
<td>$0; card and phone number at sign-up, not charged</td>
<td>VM, firewall, install</td>
<td>no end date</td>
<td>the Arm allowance is <a href="https://docs.oracle.com/en-us/iaas/Content/FreeTier/freetier_topic-Always_Free_Resources.htm">2 OCPUs and 12 GB</a>, not the 4 and 24 older guides quote; &ldquo;Out of host capacity&rdquo; on creation; instances idle for 7 days can be reclaimed</td>
</tr>
<tr>
<td><a href="https://docs.cloud.google.com/free/docs/free-cloud-features">Google Cloud free tier</a></td>
<td>$0 for one e2-micro in three US regions, 30 GB disk; card required</td>
<td>VM, firewall, install</td>
<td>no end date; the $300 trial credit lasts 90 days</td>
<td>1 GB of memory — the floor</td>
</tr>
<tr>
<td><a href="https://aws.amazon.com/free/">AWS Free Tier</a></td>
<td>$100 in credits at sign-up, up to $200</td>
<td>VM, firewall, install</td>
<td>after 6 months, or when the credits run out</td>
<td>the old always-free micro instance is gone</td>
</tr>
<tr>
<td>Cheapest VPS (<a href="https://www.hostinger.com/vps-hosting">Hostinger KVM 1</a>)</td>
<td>$6.49/mo on a 24-month term, renews at $11.99; 1 vCPU, 4 GB</td>
<td>install</td>
<td>never free; 30-day money-back</td>
<td>the renewal price</td>
</tr>
<tr>
<td><a href="https://runware.ai/free-openclaw">Runware</a></td>
<td>hosting $0, &ldquo;AI inference billed per token&rdquo;; $2 in credit with a business email, no payment to deploy</td>
<td>nothing</td>
<td>hosting stays free; tokens are paid from the first reply</td>
<td>one instance per account, <a href="https://dev.to/runware/we-hosted-openclaw-so-you-dont-have-to-3698">suspended after 5 days unused</a></td>
</tr>
<tr>
<td>KiloClaw</td>
<td>discontinued: end-of-life notice in Kilo&rsquo;s docs by <a href="https://web.archive.org/web/20260825173641/https://kilo.ai/docs/kiloclaw/overview">25 August 2026</a>, pages removed on <a href="https://github.com/Kilo-Org/kilocode/pull/14658">29 September 2026</a> as &ldquo;sunset&rdquo;</td>
<td>—</td>
<td>—</td>
<td><a href="https://vibecoding.app/blog/kiloclaw-review">reviews</a> reported $55/mo with a one-day trial from August 2026, $9/mo with 7 days before that; the managed hosts that replace it are in <a href="https://aplite.ai/blog/managed-openclaw-hosting-compared#alternatives-to-kiloclaw">Managed OpenClaw hosting compared</a></td>
</tr>
<tr>
<td><a href="https://www.hostinger.com/vps/openclaw-hosting">Hostinger managed OpenClaw</a></td>
<td>$5.99/mo on a 24-month term, $10.99 month to month, renews at $11.99; AI credits included</td>
<td>nothing</td>
<td>never free; 30-day money-back</td>
<td>no trial</td>
</tr>
<tr>
<td><a href="https://xcloud.host/openclaw-hosting/">xCloud</a></td>
<td>$9.99 the first month for 6 GB, then $19.99; your own API key</td>
<td>nothing</td>
<td>never free</td>
<td>the renewal price</td>
</tr>
<tr>
<td><a href="https://www.ampere.sh/">Ampere</a></td>
<td>the &ldquo;Free&rdquo; tier is listed at $1/mo; the FAQ says free hosting during beta, $10 in AI credits, no card</td>
<td>nothing</td>
<td>after the beta</td>
<td>the plan table and the FAQ disagree</td>
</tr>
<tr>
<td><a href="https://rapidclaw.dev/">RapidClaw</a></td>
<td>5 free messages, then $29/mo; no card to start</td>
<td>nothing</td>
<td>after 5 messages</td>
<td>—</td>
</tr>
<tr>
<td><a href="https://aplite.ai/docs/openclaw">Aplite Free</a></td>
<td>$0, no card</td>
<td>nothing — set up from the browser</td>
<td>no end date; credit packs from $5, schedules on Pro ($20/mo)</td>
<td>one agent, budget models, 25 credits a day until a pack is bought, no scheduled automations</td>
</tr>
</tbody>
</table></div>
<h2 id="the-0-path-step-by-step">The $0 path, step by step</h2>
<p>Run in a fresh container of our 2026.9.8 runtime image with a 1 GB memory cap, and checked
against the <a href="https://docs.openclaw.ai/start/getting-started">upstream docs</a>. Your install will
be newer — the gateway says so the moment it starts — but the commands are the same.</p>
<ol>
<li>
<p><strong>Install.</strong> OpenClaw needs Node 24.16 or 26.1 and newer
   (<a href="https://docs.openclaw.ai/install">install docs</a>); the installer provisions it when
   missing. On macOS, Linux or WSL2:</p>
<pre><code>curl -fsSL https://openclaw.ai/install.sh | bash
</code></pre>
<p>On Windows, in PowerShell:</p>
<pre><code>iwr -useb https://openclaw.ai/install.ps1 | iex
</code></pre>
<p>Then <code>openclaw --version</code> — ours printed <code>OpenClaw 2026.9.8 (fc23bc8)</code>.</p>
</li>
<li>
<p><strong>Have a free model ready.</strong> Local: install Ollama and pull a model
   (<code>ollama pull gemma4</code> in the <a href="https://docs.openclaw.ai/providers/ollama">upstream recipe</a>).
   Or get a key from OpenRouter or Google AI Studio. Pick Ollama with the server not running
   and onboarding stops here:</p>
<pre><code>Ollama could not be reached at http://127.0.0.1:11434.
Start or restart the Ollama server for this address.
If Ollama is not installed on that machine, download it at https://ollama.com/download
</code></pre>
</li>
<li>
<p><strong>Run onboarding.</strong></p>
<pre><code>openclaw onboard
</code></pre>
<p>Choose <em>Quick start</em>, pick the provider, paste the key. OpenClaw verifies it with a
real completion, writes <code>~/.openclaw/openclaw.json</code>, starts the gateway in the
foreground and opens the dashboard. Skip the provider and everything still installs,
but the first message fails with:</p>
<pre><code>No API key found for provider "openrouter".
Auth store: ~/.openclaw/agents/main/agent/openclaw-agent.sqlite
Configure an API key (openclaw models auth paste-api-key --provider openrouter)
| missing-provider-auth
</code></pre>
<p>That command is the fix: <code>openclaw models auth paste-api-key --provider openrouter</code>.</p>
</li>
<li>
<p><strong>Check the gateway.</strong> In a second terminal, <code>openclaw gateway status</code> prints:</p>
<pre><code>Gateway: bind=loopback (127.0.0.1), port=18789 (env/config)
Dashboard: http://127.0.0.1:18789/
Connectivity probe: ok
Listening: 127.0.0.1:18789
</code></pre>
<p><code>openclaw dashboard</code> opens the browser with a one-time pairing token in the URL. With
no provider set, the Control UI lands on <em>Model Setup</em>.</p>
</li>
</ol>
<figure class="shot">
<span class="shot-bar" aria-hidden="true"><i></i><i></i><i></i></span>
<a class="zoom" href="https://aplite.ai/assets/blog/openclaw-hosting-for-free-dashboard.webp?v=d664b4aa" target="_blank" rel="noopener"><img src="https://aplite.ai/assets/blog/openclaw-hosting-for-free-dashboard.webp?v=d664b4aa" alt="The Control UI on a fresh install: the gateway is up, no model is connected yet." loading="lazy"></a>
<figcaption>The Control UI on a fresh install: the gateway is up, no model is connected yet.</figcaption>
</figure>
<ol start="5">
<li>
<p><strong>Keep it running after you log out.</strong></p>
<pre><code>openclaw gateway install
</code></pre>
<p>This installs a systemd user unit on Linux and WSL2, a LaunchAgent on macOS, a
Scheduled Task on Windows (<a href="https://docs.openclaw.ai/gateway">gateway docs</a>). On a
server, add <code>sudo loginctl enable-linger $USER</code> so the unit survives your SSH session,
as the upstream <a href="https://docs.openclaw.ai/install/oracle">Oracle guide</a> does.</p>
</li>
</ol>
<p>Where this path ends is in the same upstream FAQ: on a laptop, &ldquo;sleep, network drops, and OS
updates can interrupt the Gateway&rdquo;, and &ldquo;use a VPS if you have had gateway disconnects
before&rdquo;. On the free clouds it is the 1 GB floor and Oracle&rsquo;s reclaim of idle instances; on
free models it is the request cap, since one agent turn with tools makes several model
requests. That is when people pay — for a $6–12 VPS plus a key, or for a managed host.</p>
<h2 id="hosted-the-same-runtime-in-three-steps">Hosted: the same runtime in three steps</h2>
<p><a href="https://aplite.ai/docs/openclaw">Aplite</a> runs a standard OpenClaw instance for you, set up from the
browser. The whole path:</p>
<ol>
<li>
<p><strong>Create an account.</strong> Email or Google, no card. The <a href="https://aplite.ai/docs/plans">Free plan</a> is one
   agent, the full workspace, budget models and 300 welcome credits that never expire.</p>
</li>
<li>
<p><strong>Deploy.</strong> Say what you want done, pick the agents, press Deploy. Under a minute later
   the workspace opens with its own OpenClaw instance — currently 2026.10.10 — in its own
   container.</p>
</li>
</ol>
<figure class="shot">
<span class="shot-bar" aria-hidden="true"><i></i><i></i><i></i></span>
<a class="zoom" href="https://aplite.ai/assets/blog/openclaw-hosting-for-free-server.webp?v=26cca59a" target="_blank" rel="noopener"><img src="https://aplite.ai/assets/blog/openclaw-hosting-for-free-server.webp?v=26cca59a" alt="A Free workspace: a standard OpenClaw runtime in its own container." loading="lazy"></a>
<figcaption>A Free workspace: a standard OpenClaw runtime in its own container.</figcaption>
</figure>
<ol start="3">
<li><strong>Ask.</strong> Nothing to install and no key to paste: models come through one metered
   account, 1 credit is 1¢, spent only when the agent calls a model, and the cost sits
   under every reply.</li>
</ol>
<figure class="shot">
<span class="shot-bar" aria-hidden="true"><i></i><i></i><i></i></span>
<a class="zoom" href="https://aplite.ai/assets/blog/openclaw-hosting-for-free-chat.webp?v=514ef2a0" target="_blank" rel="noopener"><img src="https://aplite.ai/assets/blog/openclaw-hosting-for-free-chat.webp?v=514ef2a0" alt="The agent answers in the browser with no key pasted anywhere; the line under the reply is what it cost." loading="lazy"></a>
<figcaption>The agent answers in the browser with no key pasted anywhere; the line under the reply is what it cost.</figcaption>
</figure>
<figure class="shot">
<span class="shot-bar" aria-hidden="true"><i></i><i></i><i></i></span>
<a class="zoom" href="https://aplite.ai/assets/blog/openclaw-hosting-for-free-cost.webp?v=3f4e8db2" target="_blank" rel="noopener"><img src="https://aplite.ai/assets/blog/openclaw-hosting-for-free-cost.webp?v=3f4e8db2" alt="Settings → Cost: the welcome credits, and what the week spent." loading="lazy"></a>
<figcaption>Settings → Cost: the welcome credits, and what the week spent.</figcaption>
</figure>
<p>On top of the runtime sit a markdown workspace the agent reads and writes, an approval
queue, and the agent on your phone — Telegram from a bot token, WhatsApp and WeChat by
QR; scheduled routines are on Pro ($20/mo). What
a managed instance gives up — the shell, raw <code>openclaw.json</code>, local models, day-one releases
— is listed on the <a href="https://aplite.ai/docs/openclaw">OpenClaw page</a>. Aplite is an independent service and is
not affiliated with the OpenClaw project.</p>
<h2 id="faq">FAQ</h2>
<h3 id="is-openclaw-itself-free">Is OpenClaw itself free?</h3>
<p>Yes: open-source software under the MIT license, and the
<a href="https://docs.openclaw.ai/install">install</a> costs nothing. What costs money is the machine it
runs on and the model it calls, which is what &ldquo;free hosting&rdquo; offers differ on.</p>
<h3 id="can-i-host-openclaw-for-free-forever">Can I host OpenClaw for free forever?</h3>
<p>On your own machine with a local model through Ollama, yes — as long as the machine stays
on. Oracle&rsquo;s Always Free tier has no end date but reclaims idle instances and is often out of
capacity; Runware charges nothing for hosting and bills tokens; Aplite&rsquo;s Free plan has no end
date and welcome credits that do not expire. The rest are trials or credits.</p>
<h3 id="do-i-need-a-credit-card-for-free-openclaw-hosting">Do I need a credit card for free OpenClaw hosting?</h3>
<p>Not for your own machine, Runware, RapidClaw, Ampere during its beta, or Aplite Free. Oracle
takes a card and a phone number at sign-up without charging; Google Cloud requires a payment
method; Hostinger and xCloud are paid from day one.</p>
<h3 id="do-i-need-an-api-key">Do I need an API key?</h3>
<p>Self-hosting, yes — unless the model runs locally with Ollama; the key is where the token
bill goes. On Aplite there is no key: models run through a metered account and usage is
billed in credits.</p>]]></content:encoded>
  </item>
  <item>
    <title>An AI second brain that keeps your notes up to date</title>
    <link>https://aplite.ai/blog/ai-second-brain</link>
    <guid isPermaLink="true">https://aplite.ai/blog/ai-second-brain</guid>
    <pubDate>Tue, 06 Oct 2026 00:00:00 +0000</pubDate>
    <description>A second brain usually means notes you file and an AI you ask. Here the agent reads your notes, adds to them on its own and asks before every change.</description>
    <content:encoded><![CDATA[<figure class="fig">
<a class="zoom" href="https://aplite.ai/assets/blog/ai-second-brain.webp?v=631d6b69" target="_blank" rel="noopener"><img src="https://aplite.ai/assets/blog/ai-second-brain.webp?v=631d6b69" alt="Aplite added two things from yesterday's chat to your plan — and waits for Accept." width="1560" height="820" decoding="async"></a>
<figcaption>Aplite added two things from yesterday's chat to your plan — and waits for Accept.</figcaption>
</figure>
<p>Building an AI second brain usually goes one of two ways. You install an AI plugin
for Obsidian and ask questions about your vault. Or you point Claude at a folder of
markdown and write instructions for it to tidy and connect your notes. Either way the
notes are still yours to keep up: the AI answers when asked, and the filing,
updating and remembering stay with you.</p>
<p>A second brain that actually takes work off your hands does three more things. It
adds to your notes without being asked. It runs on a schedule, so the weekly summary
is there on Friday. And it shows you every change before it lands, so you can trust
it with the folder.</p>
<p>This is how notes work in <a href="https://aplite.ai/what-is-aplite">Aplite</a>.</p>
<h2 id="plain-markdown-one-tree">Plain markdown, one tree</h2>
<p>Every workspace is a folder of markdown files: your notes, your lists, the plans for
the week — and next to them the agents&rsquo; own instructions and memory. Headings,
checklists, tables and <code>[[wikilinks]]</code> work the way you expect, and pasting markdown
in keeps it markdown. Nothing sits in a format only the app can read.</p>
<p>The agent reads the same files you do. Ask it what is left on a project and it
answers from the note, not from a guess.</p>
<h2 id="it-writes-you-accept">It writes, you accept</h2>
<p>When the agent wants to change something — two lines in today&rsquo;s plan, a new row in a
list, a whole new note — it <strong>proposes</strong> the change. The new lines appear right in the
note on a green rail with Accept and Reject next to them. You can accept, reject, or
edit the proposal first. Nothing is applied until you say so, and every proposal also
waits in <a href="https://aplite.ai/docs/approvals">Approvals</a> in the header.</p>
<p>When a task is ambiguous, it asks one question back instead of guessing: plan the
weekend outdoors or around a concert?</p>
<h2 id="it-remembers-in-the-open">It remembers in the open</h2>
<p>What the agent learns about you goes into its own memory notes — dated files in the
same tree, which you can open, read and correct. &ldquo;Film night is Friday&rdquo; is a line in a
file, not a hidden setting. If it remembered something wrong, you delete the line.</p>
<h2 id="what-to-ask-it">What to ask it</h2>
<p>Each of these is one sentence. The agent creates the task, the schedule and the note
it files into.</p>
<blockquote><p>Every Friday, check arXiv for new papers on my topics and file two-line summaries in my notes.</p></blockquote>
<p>The <a href="https://aplite.ai/use-cases/arxiv-watch">arXiv keyword watch</a> keeps a reading list you never have
to update yourself.</p>
<figure class="fig">
<a class="zoom" href="https://aplite.ai/assets/results/arxiv-watch.webp?v=385229a1" target="_blank" rel="noopener"><img src="https://aplite.ai/assets/results/arxiv-watch.webp?v=385229a1" alt="Friday's new papers, two lines each, filed into the reading note." width="1720" height="1360" decoding="async"></a>
<figcaption>Friday's new papers, two lines each, filed into the reading note.</figcaption>
</figure>
<blockquote><p>Each evening, transcribe the new audio in my workspace and file it as clean notes.</p></blockquote>
<p><a href="https://aplite.ai/use-cases/voice-memo-transcripts">Voice memo transcripts</a>: record on the go, read
tidy notes in the evening.</p>
<blockquote><p>Every Friday afternoon, turn this week&rsquo;s notes and numbers into a short deck.</p></blockquote>
<p>A <a href="https://aplite.ai/use-cases/weekly-deck">weekly status deck</a> built from what you already wrote down.</p>
<blockquote><p>Every morning at eight, send me my day on Telegram — calendar, weather, and what&rsquo;s waiting in my notes.</p></blockquote>
<p>The <a href="https://aplite.ai/use-cases/morning-briefing">morning briefing</a> reads your notes so the reminder
comes to you — see <a href="https://aplite.ai/blog/ai-assistant-in-telegram">an AI assistant in Telegram</a>.</p>
<h2 id="bringing-your-notes-in">Bringing your notes in</h2>
<p>Create notes in the workspace, or paste them in from wherever they live now — a note
pasted as markdown arrives with its headings, lists and links intact. Start with the
notes you actually work from: the current projects, the running lists. The agent gets
useful as soon as there is something real to read.</p>
<h2 id="your-notes-your-models">Your notes, your models</h2>
<p>Each workspace runs in its own container, on a server set up for you in under a
minute — no vault sync to configure, no API keys to wire into a plugin. The agent runs
on any model from the <a href="https://aplite.ai/docs/models">catalog</a>, and token usage shows under each reply. Once
you trust it with a folder — a reading list, a journal — a standing rule lets it write
there without asking; everything else still comes as a proposal.</p>]]></content:encoded>
  </item>
  <item>
    <title>An AI dashboard generator that keeps the numbers current</title>
    <link>https://aplite.ai/blog/ai-dashboard-generator</link>
    <guid isPermaLink="true">https://aplite.ai/blog/ai-dashboard-generator</guid>
    <pubDate>Tue, 06 Oct 2026 00:00:00 +0000</pubDate>
    <description>Ask for a dashboard in one sentence and get a finished page — built from your notes, refreshed on a schedule, published at a link when you want.</description>
    <content:encoded><![CDATA[<figure class="fig">
<a class="zoom" href="https://aplite.ai/assets/blog/ai-dashboard-generator.webp?v=71247e43" target="_blank" rel="noopener"><img src="https://aplite.ai/assets/blog/ai-dashboard-generator.webp?v=71247e43" alt="One sentence in, a launch dashboard out — and every morning the agent refreshes the numbers." width="1560" height="820" decoding="async"></a>
<figcaption>One sentence in, a launch dashboard out — and every morning the agent refreshes the numbers.</figcaption>
</figure>
<p>Most AI dashboard generators start from a file: upload a spreadsheet, get charts back.
That works once. The next week the numbers have moved, and you are uploading again —
because the dashboard is a picture of the data on the day you made it.</p>
<p>A dashboard worth having keeps itself current. In <a href="https://aplite.ai/what-is-aplite">Aplite</a>, the agent
that builds the page is the same one that updates it, on a schedule, from the same
places the numbers live.</p>
<h2 id="one-sentence-in">One sentence in</h2>
<blockquote><p>Make me a dashboard for my launch week.</p></blockquote>
<p>That is the whole input. Before it builds, the agent asks two or three questions — as
cards in the chat, not a form: what the page should track, how it should look.</p>
<figure class="fig">
<a class="zoom" href="https://aplite.ai/assets/blog/ai-dashboard-generator-questions.webp?v=0022187c" target="_blank" rel="noopener"><img src="https://aplite.ai/assets/blog/ai-dashboard-generator-questions.webp?v=0022187c" alt="The agent asks before it builds — a choice of look, one click to answer." width="684" height="506" decoding="async"></a>
<figcaption>The agent asks before it builds — a choice of look, one click to answer.</figcaption>
</figure>
<p>Then it builds, and the preview refreshes live while it works. About a minute later the
finished page is under <a href="https://aplite.ai/docs/pages">Pages</a> in your workspace.</p>
<h2 id="designed-not-generated">Designed, not generated</h2>
<p>The agent does not write HTML from nothing. Every page starts from one of nine finished
designs — dashboard, tracker, report, landing page, profile, mini-app, event invite,
link-in-bio, menu — and the agent fills them with your content. That is why the result
looks designed on the first try, even on an inexpensive model.</p>
<p>There are three looks — dark and calm, light and clean, bold and colorful — and &ldquo;make it
lighter&rdquo; later is one sentence, not a redesign.</p>
<figure class="fig gal">
<div class="gal-grid"><a class="live" href="https://launch-dashboard-855cf209.aplite.page/" target="_blank" rel="noopener"><img src="https://aplite.ai/assets/shots/C3-launch-dashboard.webp?v=c0deaa17" alt="Launch dashboard" loading="lazy"><span>Launch dashboard</span></a><a class="live" href="https://weekly-report-855cf209.aplite.page/" target="_blank" rel="noopener"><img src="https://aplite.ai/assets/shots/C4-weekly-report.webp?v=4daf40bb" alt="Weekly report" loading="lazy"><span>Weekly report</span></a><a class="live" href="https://run-streak-855cf209.aplite.page/" target="_blank" rel="noopener"><img src="https://aplite.ai/assets/shots/C5-run-streak.webp?v=7187ee21" alt="Run streak tracker" loading="lazy"><span>Run streak tracker</span></a><a class="live" href="https://focus-timer-855cf209.aplite.page/" target="_blank" rel="noopener"><img src="https://aplite.ai/assets/shots/C6-focus-timer.webp?v=e1a5245d" alt="Focus timer" loading="lazy"><span>Focus timer</span></a><a class="live" href="https://bakery-landing-855cf209.aplite.page/" target="_blank" rel="noopener"><img src="https://aplite.ai/assets/shots/C7-bakery-landing.webp?v=806e4239" alt="Bakery landing page" loading="lazy"><span>Bakery landing page</span></a><a class="live" href="https://cafe-menu-855cf209.aplite.page/" target="_blank" rel="noopener"><img src="https://aplite.ai/assets/shots/C8-cafe-menu.webp?v=04dd2b34" alt="Café menu" loading="lazy"><span>Café menu</span></a><a class="live" href="https://wedding-invite-855cf209.aplite.page/" target="_blank" rel="noopener"><img src="https://aplite.ai/assets/shots/C9-wedding-invite.webp?v=2d9fbb66" alt="Wedding invite" loading="lazy"><span>Wedding invite</span></a><a class="live" href="https://designer-profile-855cf209.aplite.page/" target="_blank" rel="noopener"><img src="https://aplite.ai/assets/shots/C10-designer-profile.webp?v=ca78fd9f" alt="Designer profile" loading="lazy"><span>Designer profile</span></a><a class="live" href="https://creator-links-855cf209.aplite.page/" target="_blank" rel="noopener"><img src="https://aplite.ai/assets/shots/C11-creator-links.webp?v=22943c6e" alt="Creator links" loading="lazy"><span>Creator links</span></a></div>
<figcaption>Nine pages, each built from one sentence.</figcaption>
</figure>
<h2 id="where-the-numbers-come-from">Where the numbers come from</h2>
<p>The agent builds from what it can read:</p>
<ul>
<li><strong>your notes</strong> — a launch checklist, a list of runs, a project&rsquo;s status;</li>
<li><strong>the web</strong> — prices, rates, public pages it checks for you;</li>
<li><strong>numbers you paste</strong> into the chat.</li>
</ul>
<h2 id="it-keeps-itself-current">It keeps itself current</h2>
<p>Dashboards, trackers and reports render from a small block of data inside the page.
After the build the agent offers automatic updates; say yes, and a
<a href="https://aplite.ai/docs/tasks-and-triggers">scheduled task</a> rewrites that block — every morning, every
Monday, every hour. The page stays fresh without a server behind it: the agent is the
backend.</p>
<p>That covers most &ldquo;I need a little app for this&rdquo; cases: a launch dashboard fed from your
notes, a habit tracker that logs your runs, a weekly report that writes itself.</p>
<h2 id="share-it-at-a-link">Share it at a link</h2>
<p>When the page is ready to show, <strong>Publish</strong> puts it at a public link you can send — on
the Pro and Power plans. Every page is one self-contained file with nothing loaded from
elsewhere, so a published page has nothing to break and nothing to leak. Change it in the
workspace and publish again; the link stays the same.</p>
<h2 id="what-it-is-not">What it is not</h2>
<p>It is not a builder for multi-page websites, and it does not take payments. It makes
pages — quickly — and keeps them alive. Nine finished examples, each with the one-line
ask that produced it, are on the <a href="https://aplite.ai/use-cases/build-a-page">Build a page</a> page.</p>]]></content:encoded>
  </item>
  <item>
    <title>An AI agent for daily tasks — no Zapier, no flowchart</title>
    <link>https://aplite.ai/blog/ai-agent-for-daily-tasks</link>
    <guid isPermaLink="true">https://aplite.ai/blog/ai-agent-for-daily-tasks</guid>
    <pubDate>Tue, 06 Oct 2026 00:00:00 +0000</pubDate>
    <description>The daily jobs worth handing to an AI agent, each set up in one sentence — scheduled, delivered to Telegram or email, and checked with you before anything leaves.</description>
    <content:encoded><![CDATA[<figure class="fig">
<a class="zoom" href="https://aplite.ai/assets/blog/ai-agent-for-daily-tasks.webp?v=f3625c62" target="_blank" rel="noopener"><img src="https://aplite.ai/assets/blog/ai-agent-for-daily-tasks.webp?v=f3625c62" alt="One sentence becomes a routine on today's timeline — and at eight the digest is in." width="1560" height="820" decoding="async"></a>
<figcaption>One sentence becomes a routine on today's timeline — and at eight the digest is in.</figcaption>
</figure>
<p>Lists of the best AI agents for daily tasks mostly name assistants you talk to: you
open a chat, ask, get an answer, close it. That helps with the task in front of you.
The tasks that eat a day are the other kind — the ones that come back every morning,
every Friday, every time a price moves — and those need an agent that runs without
being opened.</p>
<p>The classic answer is an automation tool: Zapier, Make, n8n. You draw the
automation — a trigger node, an action node, mapped fields, a test run — and the
drawing is what breaks when anything changes. An agent can take the sentence instead.</p>
<h2 id="a-routine-is-one-sentence">A routine is one sentence</h2>
<p>In <a href="https://aplite.ai/what-is-aplite">Aplite</a>, a recurring job is a <a href="https://aplite.ai/docs/tasks-and-triggers">task</a>: a
small markdown file with instructions in plain language and a few fields — when it
runs, which model runs it, where the result goes. You do not write the file. You say
what you want in the chat, and the agent creates the task, sets the schedule and asks
where to deliver.</p>
<p>Every run leaves a transcript you can open. Changing the routine is editing a
sentence, and every task sits on one screen with today&rsquo;s runs on a timeline.</p>
<h2 id="daily-tasks-worth-handing-over">Daily tasks worth handing over</h2>
<blockquote><p>Every morning, brief me on my topics — a few lines per story with the source link underneath.</p></blockquote>
<p>A <a href="https://aplite.ai/use-cases/topic-news-digest">news digest</a> of what was published since yesterday.</p>
<figure class="fig">
<a class="zoom" href="https://aplite.ai/assets/results/topic-news-digest.webp?v=2f3a400e" target="_blank" rel="noopener"><img src="https://aplite.ai/assets/results/topic-news-digest.webp?v=2f3a400e" alt="A few lines per story, the source underneath — at the same time every day." width="1720" height="1360" decoding="async"></a>
<figcaption>A few lines per story, the source underneath — at the same time every day.</figcaption>
</figure>
<blockquote><p>Every morning at seven, turn my Todoist into a realistic plan for the day.</p></blockquote>
<p>The <a href="https://aplite.ai/use-cases/todoist-daily-plan">daily plan from Todoist</a> weighs overdue and
due-today against how booked the day already is.</p>
<blockquote><p>At six on workdays, recap my Slack channels — decisions, my mentions, and questions nobody answered.</p></blockquote>
<p>A <a href="https://aplite.ai/use-cases/slack-eod-recap">Slack end-of-day recap</a>: what was decided, where you
were mentioned, what is still hanging.</p>
<figure class="fig">
<a class="zoom" href="https://aplite.ai/assets/results/slack-eod-recap.webp?v=bd54294d" target="_blank" rel="noopener"><img src="https://aplite.ai/assets/results/slack-eod-recap.webp?v=bd54294d" alt="Decisions, mentions and open questions from the day's channels." width="1720" height="1400" decoding="async"></a>
<figcaption>Decisions, mentions and open questions from the day's channels.</figcaption>
</figure>
<blockquote><p>Watch the price on this product page and message me when it moves: [link]</p></blockquote>
<p>The <a href="https://aplite.ai/use-cases/price-watch">price watcher</a> checks once a day and stays quiet until the
number changes.</p>
<p>More of the same shape: the <a href="https://aplite.ai/use-cases/morning-briefing">morning briefing</a>, a
<a href="https://aplite.ai/use-cases/market-alert">market threshold alert</a>, the
<a href="https://aplite.ai/use-cases/arxiv-watch">arXiv keyword watch</a> and the rest of the
<a href="https://aplite.ai/use-cases">use cases</a>, each with what it needs connected and when it runs.</p>
<h2 id="where-the-results-land">Where the results land</h2>
<p>Every task has a delivery field: <strong>Telegram</strong>, <strong>email</strong>, the <strong>chat</strong>, or a
<strong>document</strong> in your workspace. A digest lands where you read in the morning; a
weekly summary files itself into the note it belongs to. The platform sends the result
and logs the send — the agent does not have to remember to.</p>
<h2 id="started-by-an-event-not-a-clock">Started by an event, not a clock</h2>
<p>Some jobs start when something happens. A <strong>trigger</strong> gives a task a webhook URL: a form
submission, a payment or your own script calls it, and the run starts. Longer jobs
become a <strong>chain</strong> — steps that run in order, with a gate where you approve before the
next step goes.</p>
<blockquote><p>Answer messages from my resume page. Ask me first.</p></blockquote>
<p>That one sentence becomes a chain started by a webhook: the agent drafts the reply,
and it waits in Approvals until you say send.</p>
<h2 id="what-runs-on-its-own-and-what-asks-you">What runs on its own, and what asks you</h2>
<p>Reading, searching and writing a summary run on their own. Anything that would leave
the workspace — an email sent, a file changed — lands in <a href="https://aplite.ai/docs/approvals">Approvals</a>
first: approve it, edit it, or reject it. Routines you install from the
<a href="https://aplite.ai/docs/skills-and-connections">routines shelf</a> arrive switched off, with the schedule
filled in, and nothing runs until you turn them on.</p>
<h2 id="compared-with-zapier-and-n8n">Compared with Zapier and n8n</h2>
<p>Workflow tools are the right choice for high-volume plumbing between apps — every new
order into a spreadsheet, thousands of times a day, the same way each time. An agent
is better at the daily jobs that need reading and judgement: what in today&rsquo;s news
matters to you, which tasks fit in a busy day, which Slack thread needs an answer. It
also takes the change as a sentence — &ldquo;skip weekends&rdquo;, &ldquo;only if it&rsquo;s over 5 %&rdquo; —
instead of another branch in a diagram.</p>]]></content:encoded>
  </item>
  <item>
    <title>A personal AI assistant in Telegram that does the work</title>
    <link>https://aplite.ai/blog/ai-assistant-in-telegram</link>
    <guid isPermaLink="true">https://aplite.ai/blog/ai-assistant-in-telegram</guid>
    <pubDate>Tue, 06 Oct 2026 00:00:00 +0000</pubDate>
    <description>Most AI bots for Telegram are a chat window. A personal assistant reads your notes, runs routines on a schedule and messages you first.</description>
    <content:encoded><![CDATA[<figure class="fig">
<a class="zoom" href="https://aplite.ai/assets/blog/ai-assistant-in-telegram.webp?v=72f2bd25" target="_blank" rel="noopener"><img src="https://aplite.ai/assets/blog/ai-assistant-in-telegram.webp?v=72f2bd25" alt="Aplite writes first: a price that moved, the morning digest — in your Telegram." width="1560" height="820" decoding="async"></a>
<figcaption>Aplite writes first: a price that moved, the morning digest — in your Telegram.</figcaption>
</figure>
<p>Search for an AI bot for Telegram and you mostly find two things: chat windows that
forward your message to a model and send back the answer, and code on GitHub you can
host yourself if you have an evening and a server. Both answer when you ask. Neither
does anything while you are not asking.</p>
<p>The useful version works the other way round. It knows your notes and your tasks, it
runs the small jobs you gave it on a schedule, and it messages <em>you</em> — the briefing at
eight, the price that moved, the threshold that was crossed. Telegram is just where you
read it, because your phone is already open there.</p>
<p>This is what that looks like in <a href="https://aplite.ai/what-is-aplite">Aplite</a>, where Telegram is one of the
places an agent can deliver to.</p>
<h2 id="three-things-a-chat-bot-does-not-do">Three things a chat bot does not do</h2>
<p><strong>It works from your own material.</strong> Every Aplite workspace is a folder of markdown
notes the agent can read — projects, lists, the plan for the week. Ask it in Telegram
what is left on a project and it answers from the note, not from the internet.</p>
<p><strong>It writes to you first.</strong> An instruction like &ldquo;every morning at eight&rdquo; becomes a
<a href="https://aplite.ai/docs/tasks-and-triggers">task</a> with a schedule. The agent runs it, and the result
lands in your Telegram chat without you opening anything.</p>
<p><strong>It asks before it acts.</strong> Anything that changes your files or sends something on
your behalf goes through <a href="https://aplite.ai/docs/approvals">approvals</a>: you see what the agent wants to
do and decide.</p>
<h2 id="five-things-to-ask-it">Five things to ask it</h2>
<p>Each of these is one sentence. Copy it into the chat, and the agent sets up the task,
the schedule and the delivery to Telegram itself.</p>
<blockquote><p>Every morning at eight, send me my day on Telegram — calendar, weather, and what&rsquo;s waiting in my notes.</p></blockquote>
<p>The <a href="https://aplite.ai/use-cases/morning-briefing">morning briefing</a>: one message before your day starts.</p>
<figure class="fig">
<a class="zoom" href="https://aplite.ai/assets/results/morning-briefing.webp?v=aa59064b" target="_blank" rel="noopener"><img src="https://aplite.ai/assets/results/morning-briefing.webp?v=aa59064b" alt="Delivered to Telegram at 08:00 — calendar, weather, what is waiting in your notes." width="1720" height="1380" decoding="async"></a>
<figcaption>Delivered to Telegram at 08:00 — calendar, weather, what is waiting in your notes.</figcaption>
</figure>
<blockquote><p>Watch the price on this product page and message me when it moves: [link]</p></blockquote>
<p>The <a href="https://aplite.ai/use-cases/price-watch">price watcher</a> checks the page once a day and stays quiet
until the number changes.</p>
<figure class="fig">
<a class="zoom" href="https://aplite.ai/assets/results/price-watch.webp?v=9d0ae274" target="_blank" rel="noopener"><img src="https://aplite.ai/assets/results/price-watch.webp?v=9d0ae274" alt="Silent for days, then one message when the price drops." width="1720" height="1200" decoding="async"></a>
<figcaption>Silent for days, then one message when the price drops.</figcaption>
</figure>
<blockquote><p>Watch my tickers against the levels I set and message me only when one is crossed.</p></blockquote>
<p>A <a href="https://aplite.ai/use-cases/market-alert">market threshold alert</a> — your levels, not a feed of every
tick.</p>
<blockquote><p>Every morning, brief me on my topics — a few lines per story with the source link underneath.</p></blockquote>
<p>A <a href="https://aplite.ai/use-cases/topic-news-digest">news digest</a> on the subjects you pick, with links you
can follow from the phone.</p>
<blockquote><p>Every morning at seven, turn my Todoist into a realistic plan for the day.</p></blockquote>
<p>The <a href="https://aplite.ai/use-cases/todoist-daily-plan">daily plan from Todoist</a>: overdue and due-today
tasks, put in an order you can actually get through.</p>
<p>More routines like these — and what each one needs connected — are on the
<a href="https://aplite.ai/use-cases">use cases</a> page.</p>
<h2 id="setting-it-up">Setting it up</h2>
<ol>
<li><strong>Create a workspace.</strong> Sign in with Google or email and pick your agents — the
   workspace is ready in under a minute. The agent already answers in the web chat.</li>
<li><strong>Make a bot.</strong> In Telegram, open <a href="https://t.me/BotFather">@BotFather</a>, send
   <code>/newbot</code> and copy the token it gives you.</li>
<li><strong>Connect it.</strong> In Aplite, open <strong>Skills → Telegram</strong>, paste the token and press
   Validate. The workspace restarts to pick up the bot — about a minute.</li>
<li><strong>Say hello.</strong> Send <code>/start</code> to your new bot — a Telegram bot cannot message you
   until you have written to it once.</li>
<li><strong>Approve the chat.</strong> A pairing request appears under Telegram in Aplite; press
   Approve. Only chats you approve can talk to your agent.</li>
</ol>
<p>From then on, anything you ask for &ldquo;on Telegram&rdquo; goes there, and you can message the
agent from your phone like any other chat. The <a href="https://aplite.ai/docs/skills-and-connections">Skills and connections</a>
page has the details, including how to pause the channel without disconnecting it.</p>
<h2 id="your-bot-your-workspace">Your bot, your workspace</h2>
<p>If you have looked at building this yourself, the usual recipe is an n8n workflow: a
Telegram trigger node, a model node, a memory node, one more workflow per schedule. It
works, and every new routine is another diagram to wire. Here a new routine is one more
sentence.</p>
<p>The bot you create is yours: you name it, and it talks only to your workspace and to
the chats you approved. WhatsApp and WeChat connect on the same Skills → Channels screen
by scanning a QR code — the same agent, the same workspace.
Each workspace runs in its own container on the open-source
<a href="https://openclaw.ai">OpenClaw</a> runtime, so the setup that self-hosting guides spend an
evening on — server, config file, pairing — is already done. The models are your
choice too: the agent runs on any model from the <a href="https://aplite.ai/docs/models">catalog</a>, and token
usage shows under each reply.</p>]]></content:encoded>
  </item>
</channel>
</rss>
