Other tools

Aider

Aider is an OpenAI-compatible client. Configure it with a file or with environment variables; pick one.

# ~/.aider.conf.yml — or .aider.conf.yml inside the repository
openai-api-base: https://api.kumorouter.com/v1
model: openai/<model>
# the key goes in OPENAI_API_KEY, not in this file

The openai/ prefix in the model name is required, and it is not about who made the model. It tells Aider: speak the OpenAI protocol to the configured base URL. Every Kumo identifier needs it, including the models from other vendors.

Keep the key in an environment variable rather than in the file: .aider.conf.yml often lives inside a repository. Here and further down the page KUMO_API_KEY is the variable you export yourself; everything else only reads it. If you split a main and a weak model, name an identifier from the catalog for the second one too, with the same prefix.

Verify with aider --message "reply with OK" --no-auto-commits in an empty directory.

Zed

In Zed the address lives in the settings file and the key does not: the key is entered in the agent panel and stored in the operating system's keychain.

• ~/.config/zed/settings.json — user settings. On Windows that is %APPDATA%\Zed\settings.json.

Put Kumo in as the OpenAI provider's address and list the models you want in the picker. Leave the rest of your settings alone — the block is merged into them.

{
  "language_models": {
    "openai": {
      "api_url": "https://api.kumorouter.com/v1",
      "available_models": [
        { "name": "<model>", "display_name": "Kumo" }
      ]
    }
  }
}

Zed also wants the model's context window in that entry. Take it from the model's card in the catalog and put it in the field your Zed release names — key names differ between releases, so follow the docs of the one you have installed.

Paste the key in the agent panel, under the OpenAI provider. Then restart Zed: the address is read at start-up.

OpenClaw

OpenClaw is configured by a single file — ~/.openclaw/openclaw.json. The provider is added to whichever ones are already described there.

{
  "models": {
    "providers": {
      "kumo": {
        "baseUrl": "https://api.kumorouter.com/v1",
        "apiKey": "${KUMO_API_KEY}",
        "api": "openai-completions",
        "models": [{ "id": "<model>" }]
      }
    }
  }
}

api: "openai-completions" picks the dialect: OpenAI Chat Completions. The key itself is not written into the file: OpenClaw reads the environment variable only in the ${KUMO_API_KEY} form, with the braces — a plain $KUMO_API_KEY is sent to the provider as a literal string. The list can hold several models, one entry per identifier.

Any other tool

A tool missing from this list is not a tool that cannot be connected. The order is the same for all of them.

1. 1Work out the dialectWhich SDK the tool pulls in, OpenAI or Anthropic. That decides the base URL and nothing else.
2. 2Find the provider settingA file in your home directory, a project config, environment variables, or a settings screen.
3. 3Enter the address and the key`https://api.kumorouter.com/v1` for the OpenAI dialect, `https://api.kumorouter.com` for the Anthropic one.
4. 4Name a modelAn identifier from the catalog.
5. 5VerifyOne call, and one line in the console log.

If the tool cannot be pointed at a foreign address at all, that is a dead end, and admitting it costs less than hunting for a setting that does not exist.

The fastest way through this work is the coding agent already on your machine: it reads the real files instead of guessing at them. A ready-made prompt is on the integrations overview, along with the platform's live recipes — including the ones not written up separately here.

A prompt for your agent → Check your key →