ChatProvider

The port for chat model implementations.

ChatProvider is the trait every chat model adapter implements. It is the port in the hexagonal architecture; the adapt/ layer provides the concrete adapters.

The full file is src/provider/traits.rs.

API

#[async_trait]
pub trait ChatProvider: Send + Sync {
    fn id(&self) -> ProviderId;
    fn capabilities(&self) -> ProviderCapabilities;
    async fn complete(&self, request: ChatRequest) -> ProviderResult<ChatResponse>;
    async fn stream(&self, request: ChatRequest) -> ProviderResult<ChatStream> {
        Err(ProviderError::Unsupported { provider: self.id(), feature: "chat_stream".into() })
    }
}

Capabilities

pub struct ProviderCapabilities {
    pub chat: bool,
    pub chat_stream: bool,
    pub tool_calls: bool,
    pub vision: bool,
    pub structured_output: bool,
    pub context_overflow_retry: bool,
}

Error classification

ProviderError distinguishes retryable vs. non-retryable failures:

• Retryable: RateLimited, Timeout, Overloaded, Transport
• Non-retryable: Authentication, Unsupported, BadRequest, Decode

The is_retryable() and is_context_overflow() methods on ProviderError are what ModelRouter uses to decide whether to retry or fall back.

See also

• EmbeddingProvider — the embedding port.
• ProviderRegistry — the routing layer.
• OpenAI Adapter — the concrete implementation.