Rig
An open-source Rust AI agent library with local Candle models, cloud providers, structured output and vector search. MIT licensed.
Rig is a Rust library for developers building AI agents and LLM applications with local or cloud models. Its unified API covers model providers and vector stores, so applications can share the same abstractions across backends. It's open source under MIT.
For local AI, its Candle integration supports Llama, SmolLM2 and Qwen3. FastEmbed provides another embedding integration. Cloud connections include AWS Bedrock, Google Vertex AI and Google Gemini gRPC; those model calls go to external services, while Candle runs models locally. Rig is a library you incorporate into your own application, rather than a standalone chat app.
Agents can call type-safe tools, return structured output and handle multi-turn conversations with streaming responses. Completion and embedding workflows sit alongside support for transcription, audio generation and image generation, depending on the model backend. Conversation memory and history policies help manage context across turns.
For applications that search their own documents, Rig includes an in-memory vector store and integrations with Qdrant, LanceDB, PostgreSQL, Milvus and other databases. Recording and replay support lets developers capture agent execution and check replay compatibility, with checkpoints for execution state.
The portable core and classic agent runtime also support browser WebAssembly. MCP support runs only in native applications, and WASI isn't supported.
