Connect Kilo Gateway to OpenClaw with a Single API Key

A single OpenAI-compatible endpoint and API key let Kilo Gateway serve requests across a wide range of models.

Install plugin

openclaw plugins install @openclaw/kilocode-provider

If a Gateway is already running, the installation is picked up on the spot; if not, it applies at the next startup. See Apply changes and inspect.

Setup

Create an account

Open app.kilo.ai, log in or register an account, and then create an API key.

Run onboarding

openclaw onboard --auth-choice kilocode-api-key

You can also define the environment variable yourself:

export KILOCODE_API_KEY="<your-kilocode-api-key>" # pragma: allowlist secret

Verify the model is available

openclaw models list --provider kilocode

Your model entries survive onboarding, and generated catalog rows are left to discovery. When models.mode: "replace" is used, the built-in catalog is written as well, since that mode does not run discovery.

Default model and catalog

The default model is kilocode/kilo-auto/balanced, the balanced smart-routing tier of Kilo Gateway. OpenClaw publishes no task-to-upstream-model mapping for it; routing behind kilo-auto/balanced is handled by Kilo Gateway.

During startup OpenClaw calls GET https://api.kilo.ai/api/gateway/models and merges a nonempty public list with the static routing entry. Only kilocode/kilo-auto/balanced (Auto Balanced, input: ["text", "image"], reasoning: true, contextWindow: 1000000, maxTokens: 65536) is included in the static catalog.

Every model on the gateway can be addressed as kilocode/<upstream-id> (for instance kilocode/anthropic/claude-sonnet-4, kilocode/openai/gpt-5.5). To view the complete discovered list, run /models kilocode or openclaw models list --provider kilocode.

Config example

{
  env: { vars: { KILOCODE_API_KEY: "<your-kilocode-api-key>" } }, // pragma: allowlist secret
  agents: {
    defaults: {
      model: { primary: "kilocode/kilo-auto/balanced" },
    },
  },
}

Behavior notes

Transport and compatibility

Because Kilo Gateway is OpenRouter-compatible, requests follow the proxy-style OpenAI-compatible path instead of native OpenAI request shaping (there is no store and no OpenAI reasoning-effort payload).

• Kilo refs backed by Gemini remain on the proxy-Gemini path: Gemini thought signatures are sanitized there by OpenClaw, but native Gemini replay validation and bootstrap rewrites are not enabled.
• A Bearer token derived from your API key is used for requests.

Stream wrapper and reasoning

The Kilo stream wrapper attaches an X-KILOCODE-FEATURE request header (default openclaw, which the KILOCODE_FEATURE env var can override) and normalizes reasoning-effort payloads for models that support them.

Warning

Reasoning-effort injection is skipped for kilocode/kilo-auto/balanced and x-ai/* refs. If reasoning support is needed, use a concrete model ref such as kilocode/anthropic/claude-sonnet-4.

Troubleshooting

• Should model discovery fail, OpenClaw reports an unavailable catalog refresh. Static rows are not substituted for the failed request, and an empty response is not turned into kilocode/kilo-auto/balanced.
• Check that your API key is valid and that the models you want are enabled on your Kilo account.
• If Gateway runs as a daemon, that process needs KILOCODE_API_KEY to be available (in ~/.openclaw/.env or through env.shellEnv, for example).

Related

• Model selection, Picking providers, model refs, and failover behavior.

• Configuration reference, Complete OpenClaw configuration reference.

• Kilo Gateway, Kilo Gateway dashboard, API keys, and account management.