Run OpenClaw on BoostRail

OpenClaw runs agents continuously, which makes it one of the most token-hungry setups a developer can have. It also expects you to bring your own model access. Adding BoostRail as a custom provider gives every agent one key, one bill, and the full catalog to choose from.

Why a gateway suits a long-running agent

A continuously running agent repeats context constantly, which is exactly the traffic shape that caching is built for: cached input bills at 10% of the input price, cache writes carry no surcharge, and token prices sit at or below the provider's list.

It also means a single provider outage or rate wall stops your agent mid-task. Through the gateway you get one OpenAI-compatible endpoint across leading models, so switching models is a string change in your config rather than a new account, a new key, and a new integration.

Add BoostRail as a custom provider

OpenClaw reads ~/.openclaw/openclaw.json (JSON5 — comments and trailing commas are fine). Define the provider under models.providers, then reference its models as boostrail/<model-id>.

{
  models: {
    mode: "merge",
    providers: {
      boostrail: {
        baseUrl: "https://api.boostrail.com/v1",
        apiKey: "${BOOSTRAIL_API_KEY}",
        api: "openai-completions",
        models: [
          {
            id: "claude-sonnet-5",
            name: "Claude Sonnet 5",
            input: ["text"],
            contextWindow: 200000,
            maxTokens: 64000,
          },
        ],
      },
    },
  },
  agents: {
    defaults: {
      model: { primary: "boostrail/claude-sonnet-5" },
    },
  },
}

• Set BOOSTRAIL_API_KEY in the environment OpenClaw runs in. The key value goes in the environment, not in the config file.
• Add one model entry per model you want to use, with the exact id from the Models page. The catalog ids are stable strings — claude-sonnet-5, deepseek-v4-pro, gpt-5.4, and so on.
• If your setup restricts models through agents.defaults.modelPolicy.allow, add the ref boostrail/* or the exact model refs — a provider definition alone does not make a model selectable.
• contextWindow and maxTokens describe the model you picked, not the gateway. Check the Models page for the values that apply to your model, including the models that carry a context cap.

Anthropic-protocol alternative

OpenClaw also ships an anthropic-messages adapter. If you want Claude models over Anthropic's native protocol rather than the OpenAI-compatible face, point that adapter at the gateway root instead.

boostrail-anthropic: {
  baseUrl: "https://api.boostrail.com",
  apiKey: "${BOOSTRAIL_API_KEY}",
  api: "anthropic-messages",
  models: [
    { id: "claude-sonnet-5", name: "Claude Sonnet 5", input: ["text"] },
  ],
}

Confirm it is working

Before letting an agent run unattended, make one call with the same key and base URL. If curl succeeds and OpenClaw does not, the problem is the config file or the model allowlist, not the gateway.

curl https://api.boostrail.com/v1/chat/completions \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"claude-sonnet-5","messages":[{"role":"user","content":"reply with OK"}]}'

A working setup returns a JSON body with a choices array and the model's reply inside it. Your usage and the exact price applied to that request appear in the console within a few seconds.

OpenClaw FAQ

One key for every agent you run

Create a key, add the provider block, and point your agents at any model in the catalog.