> ## Documentation Index
> Fetch the complete documentation index at: https://docs.nanoclaw.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# What is NanoClaw?

> A personal AI assistant you run on your own machine and text from Telegram, WhatsApp, Discord, Slack, or the terminal — every agent working inside its own sandbox.

NanoClaw is a personal AI assistant you run on your own machine and reach from the messaging apps you already use — Telegram, WhatsApp, Discord, Slack, or just your terminal. Text it like you'd text a coworker. Each agent works inside its own locked-down sandbox, so it does real work without touching the rest of your computer.

It's self-hosted and yours — clone the repo, run one setup command, connect a chat app. Claude powers agents by default, with Codex, OpenCode, and Ollama as alternatives.

<Info>
  **New here? Four words that show up everywhere:**

  * **Agent** — one AI identity with its own memory, workspace, and personality.
  * **Channel** — a messaging app you reach an agent through (Telegram, WhatsApp, … or the terminal).
  * **Sandbox** — the isolated Docker container an agent runs in; it only sees what you explicitly share.
  * **Provider** — the AI model behind an agent (Claude by default; Codex, OpenCode, or Ollama optional).
</Info>

## What can it do?

Everything below works on a fresh install, from a plain chat message — no configuration first:

* **Research** — agents can search the web and fetch pages out of the box. *"What changed in the new EU battery regulation?"*
* **Reminders and recurring work** — *"remind me in 20 minutes"* or *"every weekday at 9, summarize my repo's new issues"*. Scheduled tasks fire back into the same chat and survive restarts.
* **Real code and files** — the agent writes and runs code inside its sandbox, and its working files persist between conversations.
* **Read what you send** — share a file or photo over a chat app and ask about it; attachments land in the sandbox for the agent to open.
* **Remember you** — ask it to remember your name, your preferences, your projects. Each agent keeps its own persistent memory.
* **Grow into a team** — your agent can spawn companion agents, each with its own sandbox and memory, and delegate work to them. See the [multi-agent swarm guide](/guides/multi-agent-swarm).

The [quickstart](/quickstart#things-to-try-first) ends with copy-paste messages that exercise each of these.

<CardGroup cols={2}>
  <Card title="Run it" icon="rocket" href="/quickstart">
    Clone, install, and talk to your first agent.
  </Card>

  <Card title="Coming from v1 or OpenClaw" icon="route" href="/migrate-from-v1">
    What changed, what carries over, and the migration skills that do the work.
  </Card>

  <Card title="Make it yours" icon="wrench" href="/extend/overview">
    Add channels, tools, skills, and alternative agent providers to your fork.
  </Card>

  <Card title="How it works" icon="diagram-project" href="/concepts/architecture">
    The entity model, session databases, and the inbox/outbox pattern in depth.
  </Card>
</CardGroup>

## Under the hood, briefly

One idea drives the whole system: every message — a chat, a webhook, a scheduled job, one agent delegating to another — is a row in a SQLite queue, and every reply leaves through another one. There's no separate scheduler, RPC layer, or job system to learn. Channels aren't bundled features — a skill like `/add-telegram` copies exactly the adapter you asked for into your install. And the whole codebase is \~226k tokens, small enough for a long-context coding agent to hold in context — which is also how you customize it: edit the code, not a config sprawl.

The full picture — routing, sessions, the container lifecycle — lives in [Architecture](/concepts/architecture); the threat model in [Security](/concepts/security).

## Source and community

* **Source**: [github.com/nanocoai/nanoclaw](https://github.com/nanocoai/nanoclaw) (MIT)
* **Discord**: [community server](https://discord.gg/VDdww8qS42)