Home

TinyAgents Wiki

TinyAgents is a Rust-native framework for building LLM agents as typed, inspectable workflows.

The repository README is intentionally marketing-level: features, diagrams, and quick setup. This wiki holds the implementation-oriented documentation, examples, provider setup, and architecture notes.

Start Here

• Quick Start: install, run the examples, and understand the crate layout.
• Examples: runnable graph, harness, provider, tool, structured output, and .rag examples.
• Providers: OpenAI, Anthropic, Ollama, DeepSeek, Groq, xAI, OpenRouter, Together, Mistral, and compatible endpoints.
• Graph Runtime: durable graph concepts, routing, state, and execution flow.
• Architecture: how the harness, graph, registry, expressive language, and REPL language fit together.

Current Surfaces

TinyAgents is organized around five surfaces:

• Harness: provider-neutral model calls, tools, middleware, structured output, usage, events, retries, streaming, and test doubles.
• Graph runtime: typed state graphs, nodes, edges, conditional routing, commands, checkpoints, interrupts, and topology export.
• Registry: named models, tools, agents, graphs, stores, middleware, and policy records used by Rust code and .rag source.
• Expressive language: .rag workflow blueprints that compile into the same graph runtime as Rust-built graphs.
• REPL language: .ragsh orchestration for inspecting and controlling harness and graph runs through registered capabilities.

Repository Links

• Main repository
• System specification
• Examples directory
• Contributing guide