Model Provider

model-provider defines a common abstraction over LLM providers and implements multiple provider adapters.

The goal is to let agent and workflow code depend on one Provider trait while supporting local models, hosted APIs, and provider-specific features.

Main Types

• Provider: async trait implemented by model providers.
• ChatRequest, ChatResponse, ChatMessage: common chat data model.
• ToolSpec, ToolCall: tool-calling model.
• ProviderCapabilities: capability metadata.
• TokenUsage: usage accounting.
• ReliableProvider: reliability wrapper.
• RouterProvider: route requests across multiple providers.

Provider Implementations

Current modules include:

• Anthropic
• Azure OpenAI
• Bedrock
• Claude Code
• Codex
• OpenAI-compatible providers
• Copilot
• Gemini
• Gemini CLI
• GLM
• Kilo Code CLI
• Ollama
• OpenAI
• OpenRouter
• Telnyx

Consumers

light-agent uses this crate to send chat requests and tool specs without hard-coding a single LLM provider.