HATCHABOT
Many purpose-built agents · on hardware you own · thinking with the AI you choose · reachable by the people you let in · through Telegram.
the open-source agent runtime · one isolated container per agent · the front door everyone already has
The problem
You already pay for a model that can do almost anything. What you get is one assistant, in one app, from one company, for one person — whose memory of you is whatever that company decides to keep.
Hatchabot agents remember. Facts, decisions, preferences — written to memory files you can read and edit. Tell it once who you are and every agent you create already knows.
Make one per job in a minute: a kitchen helper on a cheap model, a tax advisor on the best one, a scheduler that polls its own inbox. Let them consult each other. Clone the good ones.
Claude today, Gemini or a local model tomorrow — per agent, switchable live. Your agents are files and containers on your hardware. Move them, back them up. Nothing about them belongs to a provider.
Your family messages agents like contacts on Telegram. Invite by link or QR; private threads or a group room; memory shared or personal, and everyone told which.
Features
Name + one paragraph + a bot token → a running, remembering agent in a minute. Clone one you like; import a shared template; keep a launchpad of planned agents.
Claude (subscription or API), Gemini or a local model. Switch the model live. Classes put tiers of agents on the right model, source and runtime image in one place.
Gmail, Calendar, Drive and Sheets per agent — with “read but never send” if you like. Read-only or writable folders. Git repos it commits to. Per-agent secrets.
Cron tasks, event-triggered tasks where a zero-cost check decides whether to wake the model, run-now tests.
Grant an investing agent access to your tax and legal agents; it asks them mid-task — with scoped tokens, loop guards and rate limits.
Save a conversation to memory, download an agent's whole chat history, or recover the context of a thread that got reset. Nothing important is trapped in a transcript.
Fleet health, usage per agent per day, an audit timeline, a daily security posture with diffs, nightly backups with a restore drill, snapshots before every change.
Build a runtime candidate, try it on a single agent, promote it to the fleet or discard it. Rebuilds keep memory.
The app binds to your machine; reach it from anywhere over Tailscale, never a port on your router. Telegram works from anywhere regardless.
Built on OpenClaw, the open-source agent runtime — every Hatchabot agent is a stock OpenClaw gateway in its own Docker container. Hatchabot adds the layer OpenClaw deliberately leaves to you: many isolated installs, run as a fleet, by more than one person.
Compared
| Claude.ai / ChatGPT | Codex & coding agents | Hatchabot | |
|---|---|---|---|
| Shape | One assistant, one app, one person | One agent, one repo, one task at a time | A fleet of agents with jobs, for a household |
| Runs where | Their cloud | Their cloud or your terminal | Your machine, reachable over your private network |
| Memory | What the provider keeps; opaque | The repo and the session | Files you read, edit and back up; survives rebuilds and moves |
| Who can talk to it | You, in the app | You, in the terminal or PR | Anyone you invite, on Telegram |
| Choice of AI | That company's models | That company's models | Claude, Gemini, local — per agent, switchable live |
| Acts on its own | Limited scheduled tasks | When you run it | Cron and event-triggered tasks; consults other agents |
| Cost model | Per seat | Per token / per seat | One Claude Max plan powers every agent; local models cost nothing per token |
Not a replacement for the models — Hatchabot uses Claude. It replaces the app around it.
How it works
Each agent's home directory is a Docker volume: persona, memory, sessions, tools. The runtime image is shared, versioned, and swapped under an agent without touching the volume. TypeScript, ~950 tests, MIT, fourteen full audits and counting.
Get started
Make a bot. In Telegram open @BotFather, send /newbot, keep the token.
Get Claude ready. npm i -g @anthropic-ai/claude-code, log in once, then claude setup-token.
Install Hatchabot on the computer that stays on — one line in a terminal:
bash -c "$(curl -fsSL https://hatchabot.com/install.sh)"Checks git, Docker and Node 22 (offers to install what's missing), fetches the latest release, pulls the pre-built runtime image, installs the service.
Open the app at http://localhost:8080 — the wizard takes the setup token and the bot token, then creates the agent.
Say hi on Telegram. It will still remember this conversation next year.
A Linux or macOS computer that stays on, with Docker and Node 22+ — a Mac mini, a home server, an old laptop.
A Telegram account.
A Claude subscription (Pro for a couple of agents, Max for a household) — or an API key, or a local model with no account at all.
Quick start · Philosophy · Slide deck · Releases · Security