Models & config

Anturno has no built-in default provider. You add named model entries in user config, then connect and select them in the TUI.

Config shape

Edit ~/.anturno/config.json (user scope only — project config cannot define or select models):

{
  "models": [
    {
      "id": "openai",
      "displayName": "OpenAI",
      "baseUrl": "https://api.openai.com/v1",
      "modelId": "gpt-4.1",
      "apiKey": "${OPENAI_API_KEY}",
      "contextWindowTokens": 128000,
      "inputCostPerMTok": 3,
      "outputCostPerMTok": 15
    }
  ],
  "modelRoles": { "main": "openai" }
}

Field	Purpose
`id`	How you refer to it: `/connect openai`, `/model openai`
`baseUrl` / `modelId`	Endpoint and model name
`apiKey`	`${ENV_VAR}` reference, a literal (local only), or omit and supply later via `/connect`
`contextWindowTokens`	Context window for compaction and the TUI meter (default 128000)
`inputCostPerMTok` / `outputCostPerMTok`	Optional; without them `/cost` shows “cost n/a”

The supported production path speaks the OpenAI Responses API. A custom baseUrl must speak that protocol.

Why user scope only

models and modelRoles are never read from project config. A repo could otherwise redirect model calls through an attacker-controlled endpoint carrying your API key.

Connect → model → cost

/connect <id> <key> stores the key under ~/.anturno/credentials.json (mode 0600).
/model <id> sets the active model for the project.
Turns run against that entry — key resolution order: inline config value → stored credential → named env var.
/cost attributes dollars using the per-MTok rates, when configured.
Resuming a session re-adopts the model it was on.

With an empty models[], first-run tips you to add config instead of showing an empty connect picker.

Derived env key

If you omit apiKey and never run /connect, Anturno also looks for ANTURNO_MODEL_<ID>_API_KEY (uppercase id).