QAI SDK
Modular, type-safe Rust SDK for unified AI orchestration across 6+ providers with structured output, middleware, agents, and MCP.
Complete Feature Coverage
💬 Chat Generation
Unified model.generate() across all providers. System prompts, multi-turn conversations, and complete token usage stats.
⚡ Streaming
Real-time StreamPart enum covering text deltas, tool call deltas, usage, finish reasons, and errors.
🔧 Tool Calling
First-class function calling with JSON Schema tool definitions. Works across OpenAI, Anthropic, Google, and compatible endpoints.
📋 Structured Output
generate_object() forces models to return validated JSON. Supports Json mode (native) and Tool mode (fallback).
🗂️ Provider Registry
Resolve models by string: "openai:gpt-4o". Central hub for multi-provider applications.
🔗 Middleware
Composable LanguageModelMiddleware trait. Built-in: DefaultSettings, ExtractReasoning.
🤖 Universal Agent
Builder-pattern agent with iterative tool loop. Configurable max_steps and async tool handlers.
🌐 MCP Protocol
Full MCP client: tools, prompts, resources, live subscriptions. Stdio and SSE transports.
🖼️ Vision / Multimodal
Pass images via Content::Image. Supported on GPT-4o, Claude 3, and Gemini models.
📐 Embeddings
EmbeddingModel trait for text-embedding-3, text-embedding-004, and custom dimensions.
🎨 Image Generation
ImageModel trait for DALL-E 2/3 and Imagen. HD quality, custom sizes.
🔊 TTS / STT
SpeechModel for text-to-speech (tts-1-hd). TranscriptionModel for Whisper speech-to-text.