rig

Rig is a Rust library designed for developers who need to build applications and agents that rely on large language models. It offers a single, consistent API that works with more than twenty LLM providers, so you can choose or change a back‑end without altering the surrounding code. The project is open source under the MIT license and is hosted on GitHub, where it has accumulated thousands of stars and millions of downloads.

When you add Rig to a Rust project (`cargo add rig`), the crate supplies type‑safe structures for constructing prompts, setting parameters such as temperature or token limits, and handling responses. In practice, you create a request object that includes the model identifier, the text prompt, and any optional settings; the compiler checks that these fields match the expected types. After the request is sent, the library can deserialize the model's reply directly into a Rust struct if you have asked for structured output like JSON. This eliminates manual parsing and lets you work with strongly typed data throughout your application.

Rig also bundles a modular tool system that lets you define functions the model can call during a session. For example, you might expose a search function or a database lookup; the model receives a description of the tool, decides when to invoke it, and receives the result as part of the conversation flow. The library integrates with more than twenty vector‑store services, enabling retrieval‑augmented generation where a query is first matched against stored embeddings before being sent to the LLM. Typical inputs therefore include a user query, optional tool specifications, and a vector‑store index; typical outputs are either plain text, a JSON object, or a structured response that includes tool call results.

The library is aimed at teams that require production‑grade performance and want to keep their codebase in Rust. It suits projects that already use Rust for backend services, micro‑services, or edge deployments, and where compile‑time safety is a priority. It is less suitable for developers who work primarily in other languages, as Rig does not provide native bindings for Python, JavaScript, or similar ecosystems. The listing does not specify authentication details for each provider, nor does it give pricing information for the underlying LLM services; those aspects must be handled separately.

Key features