> ## Documentation Index
> Fetch the complete documentation index at: https://docs.wolffi.sh/llms.txt
> Use this file to discover all available pages before exploring further.

# config.json

> Application configuration reference

# What config.json Stores

Located at `brain/config.json`, this file stores application-level settings. It's read on startup and can be modified via the Settings UI or by editing the file directly.

## Configuration Fields

```json theme={null}
{
  "mind": {
    "provider": "deepseek",
    "model": "deepseek-v4-pro"
  },
  "providers": {
    "deepseek": {
      "apiKey": "",
      "model": "deepseek-v4-pro"
    },
    "anthropic": {
      "apiKey": "",
      "model": "claude-sonnet-4-20250514"
    },
    "openai": {
      "apiKey": "",
      "model": "gpt-4o"
    },
    "ollama": {
      "host": "http://localhost:11434",
      "model": "gemma3:4b"
    }
  },
  "language": "en"
}
```

## Field Reference

| Field                        | Description                                                         |
| ---------------------------- | ------------------------------------------------------------------- |
| `mind.provider`              | Active LLM provider: `deepseek`, `anthropic`, `openai`, or `ollama` |
| `mind.model`                 | Active model name                                                   |
| `providers.deepseek.apiKey`  | DeepSeek API key                                                    |
| `providers.anthropic.apiKey` | Anthropic API key                                                   |
| `providers.openai.apiKey`    | OpenAI API key                                                      |
| `providers.ollama.host`      | Ollama server URL (default: `http://localhost:11434`)               |
| `language`                   | UI language: `en` (English) or `ar` (Arabic)                        |

## Model Selection

Wolffish runs on an explicitly selected model — there is no automatic fallback. In Settings → Modes, you choose your **Brain** model (provider + model), and that's the model that runs. The model you pick is the one that answers; if it fails, the turn fails rather than silently routing to another provider.

Optionally, enable **orchestrator mode** to add a second **Worker** model the Brain can delegate parallel work to. See [Orchestrator Mode](/configuration/orchestrator-mode) for how that works. Even then there's no cascade — the Brain and Worker models you selected are the ones that run.

All providers are optional — you only need the one (or two) you actually select. To use Ollama, select it as your Brain model; there's no automatic fall-through to local.

<Info>
  API keys are stored in plain text in config.json. This is by design — Wolffish is a local-first app and your workspace is your own. If you version-control your workspace with git, add `brain/config.json` to `.gitignore`.
</Info>