rig

For teams building AI applications where reliability, throughput, and resource efficiency matter embedded systems, high-frequency inference services, multi-tenant platforms Rig offers a systems-programming alternative to Python-based AI frameworks without sacrificing the composability needed for complex agent architectures.

The library provides unified provider abstractions for OpenAI, Anthropic, Google Gemini, Cohere, and Perplexity, with automatic streaming support, structured output generation via JSON schema validation, and type-safe tool calling that catches integration errors at compile time rather than runtime.

Vector store integrations cover MongoDB Atlas, LanceDB, Qdrant, and in-memory stores for RAG pipeline construction. Agent and pipeline composition follows Rust's ownership model, making resource management explicit and preventing common runtime errors that affect Python AI applications.

Rig is open-source under the MIT license and maintained by 0xPlaygrounds.

It targets Rust engineers adding LLM capabilities to existing Rust services, performance-sensitive AI applications where Python's GIL and runtime overhead are limiting factors, and teams that want compile-time guarantees about AI integration correctness.

The library is actively developed with regular additions of new provider integrations and agent primitives, positioning it as the primary Rust-native option in the LLM application framework space.