Concepts

Each page is a focused take on one idea. Read top-to-bottom for a tour of the framework, or jump to whichever concept you need.

• State and reducers: typed state, per-field merge policies, what makes nodes safe to write.
• Graphs: nodes, edges, build, invoke: the four moves you make to turn a state schema into a runnable pipeline.
• Composition: conditional edges, subgraphs, projection: routing decisions, encapsulated sub-pipelines, the parent ↔ subgraph data seam.
• Middleware: wrap node dispatch with retries, timing, logging, error transformation; per-node, per-graph, per-branch, and per-fan-out-instance registration.
• Fan-out: running the same subgraph many times in parallel, results merged back deterministically.
• Parallel branches: dispatching M heterogeneous subgraphs concurrently with per-branch state schemas and middleware.
• LLMs: how LLM calls fit into nodes, structured output, multimodal content blocks, tool definitions, routing on parsed fields, errors at the LLM boundary.
• Prompts: versioned templates, composite backends, prompt-group observability propagation.
• Observability: node-boundary hooks, OTel mapping, log correlation.
• Checkpointing: save state at each node boundary, resume from a prior point, schema migration across versions.

If you're brand-new, Quickstart is the faster entry; under a minute to a running graph. Come back here when you want to know why things are shaped the way they are.