Module providers 

Provider-facing traits: ModelProvider + ToolExecutor.

This module is the boundary between “inert data the reducer produces” and “real I/O the effect runner performs.” Concrete provider adapters (Ollama, Anthropic, Gemini, OpenAI-compat) live under model/; tool implementations live under tool/.

The goal is that adding a provider or tool touches exactly one file — no observer plumbing, no dispatch wiring, no adapter- specific retry logic. The traits carry the full surface; the effect runner just looks up an impl and calls it.

Re-exports

pub use capabilities::Capabilities;

pub use ctx::ExecContext;

pub use ctx::FinalResponse;

pub use ctx::ProgressEvent;

pub use ctx::StreamContext;

pub use ctx::StreamEvent;

pub use ctx::SubagentPhase;

pub use ctx::clone_messages;

pub use ctx::test_exec_context;

pub use ctx::test_stream_context;

pub use factory::ProviderFactory;

pub use model::AnthropicProvider;

pub use model::GeminiProvider;

pub use model::ModelProvider;

pub use model::OllamaProvider;

pub use model::OpenAICompatProvider;

pub use tool::ToolExecutor;

pub use tool::ToolRegistry;

pub use tool::TuiMode;

Modules

capabilities

Per-provider capability flags.

ctx

Per-call context passed to providers and tool executors.

factory

Runtime provider construction.

model

Model adapters wrapped as ModelProvider implementations.

tool

Tool executors — one type per tool the model can call.