Orchard is a typed, concurrent language for building LLM agents. You write an agent once as a real program, run it offline with the mock provider, then point it at a real model by changing one block. Everything below teaches the language, the runtime, and the embedding API.
Browse the docs
Learn
Tutorials that teach the language end to end, from a one-line agent up.
Language Reference
Every construct: types, agent members, the agent-native primitives, and the IR.
CLI Reference
The orch command: run, check, compile, fmt, new, info, doctor, and serve.
Embedding
Drive an agent from Rust, C, Python, or WebAssembly, and inject your own boundaries.
API Reference
The Agent, Runtime, and session types, plus the injectable provider and store traits.
Standard Library
The built-in tool packs and the memory, recall, and policy primitives.
Examples
Real, annotated Orchard programs: triage, research, tool use, and scheduled agents.
FAQ
Compiled or interpreted, offline use, providers, validation, concurrency, and safety.
New here?
Start with Hello, agent, a four-line program you can run without an API key. It uses the mock provider, so there is nothing to sign up for and nothing to pay. When you are ready to run it for real, the install page has builds for macOS, Windows, and Linux, plus Homebrew, Cargo, and Docker.
Reference
Once the basics click, the language reference covers every construct in one place, and the CLI reference covers the orch command you use to check, run, compile, and serve your agents.
The complete documentation set is also published on the GitHub-hosted docs site, and the language, runtime, and examples all live in the source repository.