providers overview

how providers work

every provider implements the Provider trait. swap providers with a config change, zero code changes.

#[async_trait]
pub trait Provider: Send + Sync {
    async fn chat(&self, messages: &[Message], tools: &[Tool]) -> Result<Response>;
    fn name(&self) -> &str;
}

built-in providers

provider	config name	models
OpenRouter	openrouter	200+ models — claude, gpt-4o, llama 4, gemini 2.5, etc.
Anthropic	anthropic	claude-sonnet-4-5, claude-opus-4, claude-haiku-3-5
OpenAI	openai	gpt-4o, o3, o4-mini
Ollama	ollama	any local model
Custom	custom:https://...	any OpenAI-compatible API

custom provider examples

any service with an OpenAI-compatible API works out of the box via the custom provider — Groq, Mistral, xAI, DeepSeek, Together, Fireworks, Perplexity, Cohere, and more. see custom providers for setup examples.

reliability

all providers use:

• exponential backoff with ±25% random jitter to prevent thundering herd
• DashMap response cache with 60s TTL so identical prompts don't burn API credits
• SSRF validation on provider URLs (blocks private IP ranges)

switching providers

# via CLI flag
baihu agent -p anthropic -m "hello"

# via config
default_provider = "openrouter"

# via env var
BAIHU_PROVIDER=ollama baihu agent