OpenClaw with nRouter

Connect OpenClaw agent workflows to nRouter (https://api.nrouter.ai/v1) to govern autonomous model calls with hard budget caps, preflight guardrails, and turn-by-turn cost logs. Pointing OpenClaw to nRouter isolates risk and prevents runaway recursive loops with strict USD ceilings.

Key capabilities:

• Per-agent virtual keys: Assign distinct nRouter keys (sk-nrouter-...) to isolate budgets and telemetry across individual agent workers.
• Autonomous loop safeguards: Enforce hard USD spending caps that return HTTP 402 before runaway recursive tool calls exhaust balances.
• Server-side guardrail inspection: Pre-call prompt injection filtering scans agent inputs and tool payloads before execution.
• Drop-in OpenAI REST compatibility: Configure the standard OpenClaw provider block with nRouter's unified endpoint.

Provider Configuration

Configure the provider block in your OpenClaw settings using the OpenAI-compatible gateway specification:

{
  "models": {
    "providers": {
      "nrouter": {
        "baseUrl": "https://api.nrouter.ai/v1",
        "apiKey": "${NROUTER_API_KEY}"
      }
    }
  }
}

Export NROUTER_API_KEY in your environment, choose a model from the nRouter catalog, and run a test turn.

What nRouter adds

OpenClaw handles agent execution; nRouter governs model spend and security. Dedicated virtual keys isolate budgets per agent, preventing runaway loops from exhausting shared allowances, while full request logs record exact list-price costs and guardrail verdicts.

Agent safety checklist

1. Assign each agent a dedicated virtual key with a hard USD budget ceiling.
2. Review guardrails and cost logs in observability after staging runs.
3. For coding tools, see Claude Code with nRouter and Cursor with nRouter. For multi-agent systems, see the CrewAI guide.

Troubleshooting

If an agent turn fails (see Errors & Rate Limits):

• 401 invalid_api_key: Ensure NROUTER_API_KEY is exported in the agent's runtime environment.
• 402 insufficient_credits: Virtual key budget ceiling reached (x-nr-limit-source identifies the limit).
• 404 model_not_found: The model route does not match the active model catalog.

Frequently asked questions