Skip to main content
An open-source, self-hosted personal AI assistant. Site: https://openclaw.ai (docs: https://docs.openclaw.ai, repo: https://github.com/openclaw/openclaw).
Protocol: OpenAI-compatible
  • Base URL: https://api.orcarouter.ai/v1 (must include /v1)

​
Install

Requires Node 22+. Installs the openclaw command: 
npm install -g openclaw
Or on macOS/Linux: 
curl -fsSL https://openclaw.ai/install.sh | bash

​
Configure

Edit ~/.openclaw/openclaw.json: 
{
  "models": {
    "mode": "merge",
    "providers": {
      "orcarouter": {
        "baseUrl": "https://api.orcarouter.ai/v1",
        "apiKey": "sk-orca-...",
        "api": "openai-completions",
        "models": [ { "id": "orcarouter/auto", "name": "OrcaRouter Auto" } ]
      }
    }
  },
  "agents": { "defaults": { "model": { "primary": "orcarouter/orcarouter/auto" } } },
  "gateway": { "mode": "local" }
}

​
Run

openclaw agent --local --agent main --session-id s1 -m "Reply with exactly: OK"
Expect the reply OK.

​
Notes

  • baseUrl must live under models.providers.<name>, not inside each model entry. Putting it on a model gives Unrecognized key: "baseUrl".
  • The base URL keeps the trailing /v1. Model references use provider/model, so the auto router is orcarouter/orcarouter/auto.
  • The default agent name is main (not default). For non-interactive runs use --local with a --session-id.
  • Model names use vendor/model format. Swap orcarouter/auto for a specific model like anthropic/claude-opus-4.8 or google/gemini-2.5-flash (and update the primary reference accordingly).

​
Thinking / reasoning

OpenClaw talks to OrcaRouter over the standard OpenAI-compatible API (api: "openai-completions"), so reasoning follows the model’s own behavior — pick a reasoning-capable model (e.g. an anthropic/claude-* reasoning model) and it thinks as configured upstream.