Rust LLM apps get a unified facade—and a breaking-changes warning

What it does

Rig is a Rust library for building modular LLM-powered applications. It offers a unified interface to over 20 model providers and more than 10 vector stores, handling completions, embeddings, transcription, audio generation, and agentic workflows with multi-turn streaming. The core library compiles to WASM, and a root rig facade lets you pull in companion crates—such as rig-lancedb or rig-qdrant—only when needed via feature flags.

The interesting bit

Instead of shipping one bloated crate, Rig splits every integration into its own companion crate behind a single feature-gated facade. That keeps compile times lean and dependency trees small, which matters when you are linking vector stores like Neo4j or Milvus that you may never touch. The maintainers also openly warn that “future updates will contain breaking changes,” which is refreshingly honest for a fast-moving AI framework.

Key highlights

• Supports 20+ model providers and 10+ vector stores under one interface
• Handles streaming, transcription, audio generation, and image generation
• Core library is fully WASM compatible
• Adopted by St Jude, Neon, Nethermind, and others for production tools
• Follows the OpenTelemetry GenAI Semantic Convention for observability

Caveats

• The README explicitly warns that near-term releases will ship breaking changes, so expect migration work
• WASM compatibility applies to the core library only, not the full companion crate suite
• Some integrations (e.g., rig-onchain-kit for Solana/EVM) live outside the main repo

Verdict

Worth a look if you are building Rust-native LLM agents or RAG pipelines and want provider portability without writing your own abstraction layer. Skip it if you need a stable, v1.0 API today or if your stack is not Rust.