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

# openclaw

> Connect RemoteAgent.CHAT to a self-hosted AI gateway via the OpenClaw HTTP interface.

<Warning>
  The `openclaw` runner is in **beta** and has not been fully tested. You may encounter bugs or unexpected behavior. Use with caution.
</Warning>

The `openclaw` runner connects to a locally running OpenClaw daemon that exposes an AI gateway on a configurable port (default `18789`). OpenClaw manages its own configuration entirely via its own setup wizard — RemoteAgent.CHAT does not configure OpenClaw settings.

## Requirements

* OpenClaw installed and its daemon running before `remoteagent init`

## Setting up OpenClaw first

Before running `remoteagent init`, you must set up OpenClaw independently:

```bash theme={null}
npm install -g openclaw@latest
openclaw onboard --install-daemon
```

`openclaw onboard --install-daemon` will walk you through OpenClaw's own configuration wizard and install the background daemon. Once the daemon is running, you can proceed to initialize RemoteAgent.CHAT.

<Warning>
  Run `openclaw onboard --install-daemon` before `remoteagent init`. The
  RemoteAgent.CHAT wizard checks whether the daemon is reachable and will show
  instructions if it is not found.
</Warning>

## Setup

Once the OpenClaw daemon is running, initialize the agent:

```bash theme={null}
remoteagent init --runner openclaw
```

Or choose `openclaw` when prompted by the interactive wizard.

During `remoteagent init`, the wizard will:

1. Ask for the **gateway port** (default `18789`)
2. Check whether the OpenClaw daemon is reachable on that port
3. If the daemon is not found: display setup instructions and offer to exit or continue anyway

RemoteAgent.CHAT does not configure any OpenClaw settings — all model routing, API keys, and gateway behavior are managed through OpenClaw's own configuration.

## How it works

The runner sends POST requests to `http://localhost:{port}/v1/chat/completions` using the OpenAI API format. The port is configurable and saved in `~/.remoteagent/agents/{agentId}.json`. The prompt is sent as a user message. Response chunks are streamed via server-sent events (SSE) and forwarded to Telegram as they arrive.

Example request body sent by the runner:

```json theme={null}
{
  "model": "default",
  "messages": [
    { "role": "user", "content": "<your prompt here>" }
  ],
  "stream": true
}
```

The `model` field defaults to `"default"` — OpenClaw is responsible for routing that to the appropriate backend model.

## Use cases

* **Local model routing** — run multiple models and route based on task type
* **Cost control** — apply rate limiting or budget caps at the gateway layer
* **Custom middleware** — inject system prompts, logging, or content filters before requests reach the model
* **Air-gapped environments** — run entirely offline with no external API calls

## Pros and cons

| Pros                                       | Cons                                                    |
| ------------------------------------------ | ------------------------------------------------------- |
| Full control over the AI backend           | Requires running and maintaining the OpenClaw daemon    |
| Works with any OpenAI-compatible model     | OpenClaw daemon must be running before the agent starts |
| Air-gap compatible                         | More setup complexity                                   |
| No API key managed by RemoteAgent.CHAT     | —                                                       |
| OpenClaw handles all its own configuration | —                                                       |