> ## Documentation index
> Fetch the complete index at: https://maritime.sh/llms.txt
> Setup procedure for AI agents (install, sign in, deploy, verify): https://maritime.sh/agents.md
> This page as HTML: https://maritime.sh/docs/frameworks/zeroclaw

# ZeroClaw Guide

ZeroClaw is a single-binary Rust agent that starts in under a second, and Maritime runs its official image as a
managed agent.

*Screen recording: Recorded live on Maritime: deploy ZeroClaw from a template, open its dashboard, and chat with it,
from the first click to the first reply.*

## Overview

[ZeroClaw](https://github.com/zeroclaw-labs/zeroclaw) is a single Rust binary that exposes a gateway plus an
OpenAI-compatible chat API. It uses a fraction of the memory of Node.js-based agents (typically <5 MB resident) and
starts in under a second, which makes it a good pick for running many agents at once.

## Docker Image

Image: `ghcr.io/zeroclaw-labs/zeroclaw:v0.8.5-debian`

Debian-based runtime, gateway on port 42617, HOME is `/zeroclaw-data`.

## How It Works

Maritime configures the daemon through its native `ZEROCLAW_*` environment variables, never by editing
`config.toml`. The daemon (a) binds `0.0.0.0` with `allow_public_bind = true`, (b) has device pairing disabled so
the Maritime proxy can connect directly, and (c) runs with full autonomy: the command allowlist, approval prompts,
workspace confinement and hourly action cap are off, because the micro-VM is the sandbox. Set any `ZEROCLAW_*`
variable on the agent to override one setting. The WebSocket sub-protocol is `zeroclaw.v1`.

```
# The container runs with:
ZEROCLAW_HEADLESS=true
ZEROCLAW_ALLOW_PUBLIC_BIND=true
ZEROCLAW_GATEWAY_PORT=42617
ZEROCLAW_GATEWAY_PASSWORD=<auto-generated>
ZEROCLAW_WORKSPACE=/zeroclaw-data/workspace
```

## Environment Variables

ZeroClaw uses an OpenAI-compatible provider. If you don't supply `OPENAI_API_KEY` / `ANTHROPIC_API_KEY`, Maritime
injects a per-agent proxy token pointed at `https://api.maritime.sh/api/llm/v1`.

## Deploy

In the dashboard, open **New agent** and pick the **ZeroClaw Agent** template, or:

```
maritime create my-agent --template zeroclaw
```

## Talk to it

Chat from the terminal. The reply is a real model response, routed through the Maritime LLM proxy:

*Terminal recording: Shows a ZeroClaw agent active on serverless Firecracker, then sends it a chat message that
returns a live reply. Recorded against production.*

Or open the agent in the dashboard and click **ZeroClaw Dashboard** to load ZeroClaw's own web UI, then chat there.
Same agent, same reply, no terminal required:

*Screen recording: Clicks the ZeroClaw Dashboard button to load the native web UI (Connected, provider
openai.default), sends a question, and the agent answers in the browser. Recorded against production.*

### Upstream links & resources

- [zeroclaw-labs/zeroclaw](https://github.com/zeroclaw-labs/zeroclaw): the source repository. ZeroClaw has no separate
  docs site; the README and repo docs are the authoritative reference for its config file (`config.toml`),
  providers, and gateway protocol.
- [ghcr.io/zeroclaw-labs/zeroclaw](https://github.com/orgs/zeroclaw-labs/packages/container/package/zeroclaw): the
  exact container image Maritime deploys (the `debian` tag), with all published variants.