minirouter.shGet started →

OpenClaw + minirouter

Draft provider block for registering minirouter in OpenClaw's config, plus the spend controls an unattended agent needs.

Base URLhttps://api.minirouter.sh/v1
Get a keyhttps://minirouter.sh/key
Get a key — no account

Settings

config.json5 in the OpenClaw config directory

In order

  1. models.providers.minirouter.baseUrlhttps://api.minirouter.sh/v1

    Complete, including /v1. OpenClaw treats this as the OpenAI-compatible root and appends /chat/completions itself.

  2. models.providers.minirouter.apiopenai-completions
  3. models.providers.minirouter.apiKey${MINIROUTER_KEY}

    Read from the environment rather than written into the config file.

  4. agents.defaults.model.primaryminirouter/meta/llama-3.3-70b

    Defining the provider is not enough — the model also has to be reachable from the agent defaults, or OpenClaw declines to use it.

config.json5 — unverified draft
{
  agents: {
    defaults: {
      model: { primary: "minirouter/meta/llama-3.3-70b" }
    }
  },
  models: {
    providers: {
      minirouter: {
        baseUrl: "https://api.minirouter.sh/v1",
        apiKey: "${MINIROUTER_KEY}",
        api: "openai-completions",
        timeoutSeconds: 300,
        // cost is OpenClaw's own accounting field, in USD per million
        // tokens. Copy the current rates for each model you add from
        // https://minirouter.sh/models rather than leaving these stale.
        models: [
          {
            id: "meta/llama-3.3-70b",
            name: "Llama 3.3 70B (minirouter)",
            reasoning: false,
            input: ["text"],
            cost: { input: 0.756, output: 0.756 },
            contextWindow: 128000,
            maxTokens: 8192
          }
        ]
      }
    }
  }
}

Draft candidates · 4 models

All models →

Only rows confirmed in the current live catalog are linked. A plain row is a draft recommendation, not a claim that the model is currently available. Paste ids exactly as written, author/name.

When it fails

  • SymptomProvider loads but every model call 404s
    Codeinvalid_request

    Paste https://api.minirouter.sh/v1 complete, including the /v1. These clients treat the base URL as the OpenAI-compatible root and append /chat/completions themselves, so trimming the /v1 leaves them calling a path that does not exist. docs/errors#invalid_request →

  • SymptomProvider is defined but the model is refused
    Codeclient-side

    Declaring a provider does not make its models usable. The model has to be reachable from agents.defaults as well — check the primary model id matches the provider name and the model id exactly, provider/author/name.

  • SymptomCredits gone overnight with nothing to show
    Codeinsufficient_credits

    An agent in a retry loop spends at machine speed. Set dailyLimitNano on the key the agent uses — one key per agent, so a runaway is contained to that key rather than the whole balance. docs/errors#insufficient_credits →

  • Symptom401 on every request
    Codeinvalid_api_key

    Wrong or rotated key. Keys start with mr-live-; after a rotation the old key keeps working for 24 hours, then dies. docs/errors#invalid_api_key →

  • SymptomThe agent goes quiet mid-conversation in a chat channel
    Codeinsufficient_credits

    The balance reached $0 mid-stream. The stream ends with a terminal error naming the exact charge for tokens already delivered — but you are reading it inside WhatsApp or Telegram, where there is no UI but the message itself, which is why that error carries a URL. docs/errors#insufficient_credits →

Every API error string carries a URL resolving to /docs/errors.

Other always-on personal agents

Why this workload bills the way it does, what it costs to contain, and which models suit it: /use/personal-agents.

Agent setup: /setup · API reference: /docs