Artificial Intelligence · Open source · Rust

Build AI agents
in Rust

A modular, high-performance library for LLM apps and agents. One unified API across 20+ providers, type-safe tools and structured output — production-ready.

Get Started Star on GitHub

Try Ryzome

$cargo add rig

7K+GitHub stars

1.4M+Downloads

226+Contributors

76K+ stars across 181+ repos

agent.rs

use rig::{completion::Prompt, providers::openai};

let openai = openai::Client::from_env()?;

let agent = openai.agent("gpt-5.5")
    .preamble("You are a helpful assistant.")
    .build();

let response = agent
    .prompt("What is Rust?")
    .await?;

output

Trusted by teams shipping AI in production

Quick install

Add Rig in one line

Drop it into any Cargo project. Ship it as a single binary or compile to WASM for server, edge, and browser environments.

Then read the quickstart

install

bash$cargo add rig

Developer Experience · Agent Stack

Simple, powerful APIs

One crate takes you from a single completion to tools, streaming, RAG, and multi-agent workflows — all with the same clean Rust builder pattern.

basic_agent.rs

use rig::{completion::Prompt, client::{CompletionClient, ProviderClient},providers::openai}; let agent = openai::Client::from_env()    .agent("gpt-5")    .preamble("You are a helpful assistant.")    .build(); let response = agent    .prompt("Explain quantum computing")    .await?; println!(response);

01 · Agent

Start with one fluent builder

Create a provider client, attach a preamble, build the agent, and prompt it. The same shape scales from a tiny prototype to a production service.

Provider client from environment
Typed Rust errors with async/await
No framework lock-in

02 · Tools

Expose typed Rust functions

Register tools directly on the agent so the model can call real application capabilities without leaving Rust’s type system. Tool arguments and structured output are ordinary Rust types checked at compile time.

Macro-generated tool schema
Compile-time checked inputs
Composable with extractors

03 · Streaming

Stream tokens as they arrive

Switch to streaming when latency matters. Rig keeps the agent API familiar while giving you chunk-level control over the response.

Backpressure-friendly streams
Works with agent prompts
Ideal for chat UIs and CLIs

04 · RAG

Ground agents in your data

Attach a vector index as dynamic context and let the agent retrieve relevant documents at prompt time.

Provider-agnostic embeddings
Pluggable vector stores
Context injected per request

05 · Workflows

Compose the whole agent stack

The same builder pattern stretches into multi-agent workflows: route work, fan out specialized agents, and merge results without leaving Rig’s primitives. Typed errors, tracing hooks, mock models, and VCR cassette tests help keep production systems observable and deterministic.