What is Aplite?

Aplite is a personal AI workspace: a team of agents on a server of your own, set up in about a minute.
Your notes, the agents’ instructions and their memory are markdown files in one tree, Obsidian-style; every change an agent makes waits for your approval.
Ask in plain words: the agents build pages you can publish and set up routines that deliver to Telegram, email or a page.
Skills, apps and connections switch on with a click; under the hood, OpenClaw runs any model on OpenRouter, metered in credits.
What the agents learn, they write down in their own notes and read next time.

One-minute server setup

Pick the agents and a model for each, the apps and the connections, then size the server and choose a region. Deploy: the server comes up, the apps install, the agents move in, the workspace opens — about a minute on the clock — and the first agent asks what to start with.

Pick your team, size your server, Deploy — a minute later Atlas is online and asking what to start with

It reads your files, runs your workspace, makes its own notes

Everything is a file in one tree — your notes, the agents’ instructions, their memory. The agent reads what is there and works in the open: a line added to your digest, a row in a list, a whole new note, each a proposal you accept or reject in place. When an ask needs a choice, it asks one question back. What it learns goes into its own memory file for next time.

Proposals in the digest and the list, a new note, one question back — and a line written to the agent's memory

Agents build your workspace

Ask for a dashboard from your notes, a workout page, a resume. Each ask ends in a link; open it and the page is there — three asks in one chat, three pages under Pages. Publish is one more click: the page goes live at a public link you can copy.

Three asks, three pages — This week, Workout, Resume — and the resume published at a public link

Routines that deliver

The day’s routines on one line — what ran, what is next, where each one delivers. At 08:00 the news digest fires and lands on Telegram. One sentence — “Answer messages from my resume page, ask me first” — becomes a chain on a webhook: read, research, draft, your approval, send. Every routine runs on the model you choose. Fifteen ready recipes in Use cases.

08:00 — the digest lands on Telegram; one sentence sets up a chain that drafts a reply and waits for your approval to send

Skills, apps and connections

What the agent can do is a list you switch on. A real browser finds a price. A spreadsheet gets read and charted. An image is generated for a note’s banner, a run video is cut in half and sent to Telegram, a reply is spoken aloud; Notion and Google Calendar answer from your data. Under “…more”: 66 built-ins, plugins, skills, apps and connections — and any MCP server you add.

Seven abilities, one demo each — and the whole catalog behind “…more”

Pages: dashboards, trackers, small apps

Nine asks, nine pages: a launch dashboard, a weekly report, a run streak, a focus timer, a bakery, a café menu, a wedding invite, a profile, a links page. The agent picks one of nine designs and fills it from your notes; a scheduled task keeps it current; publishing to a .aplite.page link is your click. Ten live examples: Build a page.

Any model, metered to the cent

GPT, Claude, Gemini, DeepSeek and the rest of OpenRouter on one account. Models sit on a ladder — the top one is primary, the rest fall back — and any agent can pin its own. Add Claude, move it to the top: the next message runs on it. Spend is metered in credits and shown by day, by agent and by model.

Claude added and moved to the top of the ladder; the spend by day, by agent and by model

Try it in your workspace

Get started