Configuration

Talon uses a JSON configuration file to manage AI providers, models, permissions, and communication channels. This guide covers all configuration options and how to set them up.

Config File Location

The talon.json configuration file is stored in your application data directory. The exact location depends on your operating system:

macOS

~/Library/Application Support/com.talon.app/talon.json

Windows

%APPDATA%\com.talon.app\talon.json

Linux

~/.config/com.talon.app/talon.json

Tip

You can create the talon.json file manually in the appropriate directory for your OS, or it will be created automatically on first launch with default settings.

Configuration Structure

The talon.json file follows this basic structure:

{
  "models": {
    "providers": [
      // Provider configurations
    ],
    "default_model": "provider/model"
  },
  "default_temperature": 0.7,
  "permission_mode": "allow",
  "channels": {
    "slack": {
      "accounts": {}
    },
    "discord": {
      "accounts": {}
    }
  }
}

AI Providers

Configure one or more AI providers under the models.providers array. Each provider requires a name and API key, with optional base URL configuration.

Provider Configuration

Each provider object requires:

• name: Provider identifier (e.g., openai, anthropic, groq)
• api_key: Your API key for the provider
• base_url (optional): Custom endpoint URL (useful for self-hosted or proxy setups)
• model (optional): Default model for this provider

Supported Providers

Talon supports the following AI providers:

• OpenAI - GPT-4, GPT-4 Turbo, GPT-3.5 Turbo
• Anthropic - Claude 3 Opus, Claude 3 Sonnet, Claude 3 Haiku
• Google Gemini - Gemini Pro and other variants
• Groq - Fast inference with Mixtral and Llama
• Together - Open-source models and fine-tuned variants
• Ollama - Local LLM inference
• OpenAI-Compatible - Any API implementing the OpenAI API specification

Example: Multiple Providers

{
  "models": {
    "providers": [
      {
        "name": "openai",
        "api_key": "sk-...",
        "model": "gpt-4"
      },
      {
        "name": "anthropic",
        "api_key": "sk-ant-...",
        "model": "claude-3-opus-20240229"
      },
      {
        "name": "groq",
        "api_key": "gsk_...",
        "model": "mixtral-8x7b-32768"
      }
    ],
    "default_model": "anthropic/claude-3-opus-20240229"
  }
}

Setting the Default Model

Specify the default model using the models.default_model field in the format provider/model:

{
  "models": {
    "default_model": "anthropic/claude-3-opus-20240229"
  }
}

This model will be used for all requests unless explicitly overridden in a channel or command configuration.

Local Models with Ollama

For self-hosted models using Ollama, configure a custom base URL:

{
  "models": {
    "providers": [
      {
        "name": "ollama",
        "base_url": "http://localhost:11434/v1",
        "api_key": "ollama"
      }
    ],
    "default_model": "ollama/mistral"
  }
}

Tip

Ollama runs locally on your machine, so inference stays private and doesn’t require internet connectivity.

Custom OpenAI-Compatible APIs

If you’re using a proxy, gateway, or custom API that implements the OpenAI specification:

{
  "models": {
    "providers": [
      {
        "name": "custom-api",
        "api_key": "your-api-key",
        "base_url": "https://your-api.example.com/v1"
      }
    ]
  }
}

Default Temperature

The default_temperature setting controls the randomness and creativity of AI responses across all models. This setting applies globally unless overridden per-channel or per-command.

Temperature Range

• 0.0: Deterministic, focused, and consistent responses (best for precise tasks)
• 0.7: Balanced (default - good for general use)
• 2.0: Maximum randomness and creativity (best for brainstorming)

Example Configuration

{
  "default_temperature": 0.5
}

Caution

Setting temperature too high (above 1.5) may result in incoherent or unpredictable outputs. For most use cases, keep it between 0.3 and 1.0.

Permission Modes

Permission modes control how Talon handles requests and determine whether actions execute automatically or require approval.

Available Modes

• plan: Talon shows a plan of what it will do and waits for confirmation before executing
• ask: Talon asks before taking each action
• allow: Talon executes actions automatically (default)
• bypass: Talon bypasses all safety checks and permission prompts

Configuration

{
  "permission_mode": "allow"
}

Permission Mode Behaviors

Mode	Behavior
plan	Shows overall plan, requires approval before any execution
ask	Requests confirmation before each individual action
allow	Executes automatically, no prompts (default)
bypass	Skips all checks and confirmations

Caution

Use bypass mode only in controlled, trusted environments. In this mode, Talon will execute commands without any safety checks or approval steps.

Channels

Talon can integrate with multiple communication platforms. Channels are configured under the channels object, organized by type (e.g., slack, discord). Each channel type has its own required fields and authentication methods.

Channel Structure

{
  "channels": {
    "slack": {
      "accounts": {
        "workspace-name": {
          // Slack-specific configuration
        }
      }
    },
    "discord": {
      "accounts": {
        "server-name": {
          // Discord-specific configuration
        }
      }
    }
  }
}

Configuring a Channel

Each channel account configuration varies by platform. For detailed setup instructions specific to each channel type, refer to the Channels documentation.

Example: Slack Integration

{
  "channels": {
    "slack": {
      "accounts": {
        "my-workspace": {
          "bot_token": "xoxb-...",
          "signing_secret": "..."
        }
      }
    }
  }
}

Tip

Different channels support different features. Check the channel-specific documentation to see what’s available for each platform.

Complete Example Configuration

Here’s a complete example with multiple providers, custom settings, and channel configuration:

{
  "models": {
    "providers": [
      {
        "name": "anthropic",
        "api_key": "sk-ant-...",
        "model": "claude-3-opus-20240229"
      },
      {
        "name": "openai",
        "api_key": "sk-...",
        "model": "gpt-4"
      },
      {
        "name": "ollama",
        "base_url": "http://localhost:11434/v1",
        "api_key": "ollama"
      }
    ],
    "default_model": "anthropic/claude-3-opus-20240229"
  },
  "default_temperature": 0.7,
  "permission_mode": "allow",
  "channels": {
    "slack": {
      "accounts": {
        "engineering-team": {
          "bot_token": "xoxb-...",
          "signing_secret": "..."
        }
      }
    },
    "discord": {
      "accounts": {
        "dev-server": {
          "bot_token": "..."
        }
      }
    }
  }
}

Next Steps

• Channel Setup: Learn how to configure Slack, Discord, and other channels
• Models & Providers: Explore available AI models and their capabilities
• Security: Review security best practices for storing API keys