Core Concepts

Rig is built from a small set of composable pieces. Once you see how they layer, everything else is a variation on the same pattern. This page is the map; each piece has its own page.

The mental model

The building blocks stack from the bottom up:

Provider clients hold a vendor’s configuration (OpenAI, Anthropic, Gemini, Ollama, and more) and a transport.
Models come from a client: completion models for text generation, embedding models for vectors, plus transcription, image, and audio models where the provider has them. You can call a model directly with a CompletionRequest.
Agents wrap a completion model with a preamble, tools, and context, and run the tool-calling loop for you. Each prompt is an AgentRunner you can tune per run; hooks observe or steer the loop.
Tools, RAG, and memory extend an agent: tools let it call your code, RAG pulls in relevant documents from a vector store, and memory loads and saves conversation history.
Workflows compose all of the above into multi-step programs with plain Rust.

flowchart TD
    Client["Provider client"] --> Model["Model"]
    Model --> Agent["Agent"]
    Tools["Tools"] -.-> Agent
    RAG["RAG<br/>(Embeddings + Vector Store)"] -.-> Agent
    Memory["Memory"] -.-> Agent
    Agent --> Output["Text or structured output"]

The Quickstart builds the first three layers (provider client, model, agent) as a complete program.

Which piece do I need?

Start from the task, not the type.

I want to…	Page
connect to a provider	Providers & Clients
send one request without the agent loop	Completions
let the model call tools over several turns	Agents
configure one agent run	AgentRunner
observe or steer the loop	Hooks
step, pause for approval, or resume a run	Durable runs
let the model call my code	Tools
get typed data out of text	Structured Output
stream tokens as they arrive	Streaming
hold a conversation	Memory
embed text	Embeddings
answer from my documents	Vector Stores & RAG
load files (text, PDF, EPUB)	Loaders
generate images, audio, or transcripts	Media
orchestrate known steps in code	Workflows
coordinate several agents	Multi-agent systems
trace runs and token usage	Observability
record a run and replay it offline	Recording & Replay
test without a live model	Testing
handle and retry failures	Error Handling