Providers & Model Configuration 🧠

MalikClaw uses a protocol-first model abstraction layer (pkg/providers). This allows seamless integration across dozens of cloud and local LLM providers using a unified configuration structure.

📋 The model_list Configuration System

All models are declared inside the model_list array within ~/.malikclaw/config.json. Each entry maps a friendly alias (model_name) to a specific model protocol and provider credential.

Core Configuration Schema

| Field | Type | Description | |-------|------|-------------| | model_name | String | Unique friendly identifier used across agents (e.g. gpt-4o, claude-3-5-sonnet). | | model | String | Protocol prefix and remote model ID (e.g. openai/gpt-4o, antigravity/gemini-2.0-flash). | | api_key | String | API key or token credential for the designated provider. | | api_base | String | (Optional) Custom endpoint base URL (e.g. local Ollama or OpenAI reverse proxy). | | auth_method | String | (Optional) Set to oauth or token for session-based authentication. |

🌐 Supported Model Protocols & Drivers

MalikClaw provides native drivers for major AI providers and generic OpenAI-compatible APIs:

| Protocol Prefix | Target Provider | Base URL / Driver | |-----------------|-----------------|-------------------| | openai/ | OpenAI | https://api.openai.com/v1 | | anthropic/ | Anthropic (SDK) | https://api.anthropic.com/v1 | | anthropic-messages/ | Anthropic (Native Messages API) | https://api.anthropic.com | | gemini/ | Google Gemini API | https://generativelanguage.googleapis.com/v1beta | | antigravity/ | Google Cloud Code Assist / AGY | Internal OAuth Client | | groq/ | Groq LPU Inference | https://api.groq.com/openai/v1 | | deepseek/ | DeepSeek API | https://api.deepseek.com/v1 | | openrouter/ | OpenRouter Unified API | https://openrouter.ai/api/v1 | | ollama/ | Ollama (Local Models) | http://localhost:11434/v1 | | zhipu/ | Zhipu AI (GLM-4) | https://open.bigmodel.cn/api/paas/v4 | | modelscope/ | ModelScope Inference | https://api-inference.modelscope.cn/v1 |

🦅 Antigravity (Google Cloud Code Assist)

Antigravity provides direct access to Google Cloud's AI models (including Gemini 2.0 Flash / Pro) via OAuth credentials:

{
  "model_name": "gemini-flash",
  "model": "antigravity/gemini-2.0-flash",
  "auth_method": "oauth"
}

Authentication: Authenticate by running:

malikclaw auth login --provider google-antigravity

⚡ High Availability & Model Fallback Chains

MalikClaw automatically fails over to backup models if a primary provider experiences downtime, rate limits, or context window limits:

{
  "agents": {
    "defaults": {
      "model": "primary-gpt-4o",
      "fallbacks": ["secondary-claude-sonnet", "local-ollama-deepseek"]
    }
  }
}

🤖 GitHub Copilot Integration

Hook directly into an active GitHub Copilot subscription without extra API keys:

{
  "model_name": "copilot-gpt4",
  "model": "github-copilot/gpt-4o",
  "connect_mode": "stdio",
  "api_base": "copilot"
}

🛠️ Advanced Model Settings

• thinking_level: Controls Anthropic reasoning depth (low, medium, high).
• max_tokens_field: Custom override for token limits (e.g. max_completion_tokens).
• request_timeout: Timeout in seconds before triggering fallback models (default: 60s).