Crate rtb_ai 

Unified AI client.

Wraps genai for the multi-provider mainstream (OpenAI / Gemini / Ollama / OpenAI-compatible) and drops down to a direct reqwest-on-Anthropic-Messages path for features genai does not yet surface — prompt caching, extended thinking, citations.

Structured output uses schemars::JsonSchema on caller-supplied types: the schema is sent with the request, and the response is validated with jsonschema before deserialising.

See docs/development/specs/2026-05-01-rtb-ai-v0.1.md for the authoritative contract.

Lint exception

Crate-level deny(unsafe_code) (not forbid) so the genai-key shim in client can locally allow(unsafe_code) the std::env::set_var it needs to hand the API key to genai. No hand-rolled unsafe blocks anywhere else.

Re-exports

pub use client::AiClient;

pub use client::ChatRequest;

pub use client::ChatResponse;

pub use client::ChatStream;

pub use client::ChatStreamEvent;

pub use config::validate_base_url;

pub use config::Config;

pub use config::Provider;

pub use error::AiError;

pub use message::Citation;

pub use message::ContentBlock;

pub use message::Message;

pub use message::Role;

pub use message::Usage;

pub use thinking::ThinkingMode;

Modules

client

AiClient — typed façade over genai + the Anthropic-direct path.

config

Config + Provider + base-URL validation.

error

Typed errors for the AI client. Every String payload runs through rtb_redact::string so leaked URLs / tokens / headers in upstream provider errors never reach our telemetry.

message

Provider-agnostic chat-message types.

thinking

Anthropic extended-thinking configuration.