# OpenClaw

Use Aurora as the inference provider for an [OpenClaw](https://openclaw.ai/)
assistant. Configure a custom provider and select its model as the agent default.

> Available Inference endpoint: `https://ai.aur.lu/v1` (Aurora).

## Install and prepare

Follow the [OpenClaw getting-started guide](https://docs.openclaw.ai/start/getting-started)
to install and initialize the gateway. Create an [Aurora token](https://docs.aur.lu/docs/inference/getting-started).
For an interactive process:

```bash
export AURORA_API_TOKEN="<your-token>"
```

If the gateway runs as a service, put the variable in its service environment or
OpenClaw's `~/.openclaw/.env` file instead:

```dotenv
AURORA_API_TOKEN=<your-token>
```

## Configure the provider and default model

Merge this into `~/.openclaw/openclaw.json`:

```json
{
  "models": {
    "mode": "merge",
    "providers": {
      "aurora": {
        "baseUrl": "https://ai.aur.lu/v1",
        "apiKey": "${AURORA_API_TOKEN}",
        "api": "openai-completions",
        "models": [
          {
            "id": "deepseek/deepseek-v4-flash",
            "name": "DeepSeek V4 Flash (Aurora)",
            "input": ["text"]
          }
        ]
      }
    }
  },
  "agents": {
    "defaults": {
      "model": {
        "primary": "aurora/deepseek/deepseek-v4-flash"
      }
    }
  }
}
```

`openai-completions` selects Chat Completions. OpenClaw uses the first `aurora/`
segment for routing and sends `deepseek/deepseek-v4-flash` to Aurora.
Registering a provider alone does not change the default model; both sections matter.

For long conversations, configure model-specific `contextWindow` and `maxTokens`
from the selected model's limits. OpenClaw's fallback budgets and zero-valued
custom-provider cost estimates are not Aurora limits or pricing.

## Start and verify

For an installed gateway service, restart it to load the credential, then check
the model configuration:

```bash
openclaw gateway restart
openclaw models list
openclaw models status
openclaw tui
```

If you run the gateway in the foreground, restart that process instead. Start a
new chat in the TUI and confirm the Aurora model is selected. Ask it to say hello,
then ask:

```text
Read AGENTS.md in your workspace and summarize it in three bullets. Do not edit files.
```

Substitute another existing workspace file if needed. Expect a tool result and
an answer based on the file. OpenClaw handles streaming in its chat interface.

## Troubleshooting

- **Missing environment variable:** A gateway service does not inherit exports
  from your current terminal. Configure its environment and restart it.
- **Another model still answers:** Check agent-specific overrides, existing
  sessions, and fallback models in addition to `agents.defaults.model.primary`.
- **Provider exists but model is missing:** Include the model in
  `models.providers.aurora.models`; an alias alone does not register it.
- **Text works, other features fail:** Web search, embeddings, audio, and image
  generation have their own configuration. This connects the main text model.

See [shared troubleshooting](https://docs.aur.lu/docs/inference/integrations#troubleshooting) and OpenClaw's
[custom provider reference](https://docs.openclaw.ai/concepts/model-providers/custom-providers).