Tutorials

Hands-on lessons that take you from a single-node graph to a running stargraph serve with replay, retrieval, classical ML, and human-in-the-loop gates wired up. Every tutorial is self-contained: paste the code, run the command, verify the output.

Learning path

Beginner — graph mechanics

• First graph — a two-node graph with a Pydantic state model, routed by a single rule, executed via stargraph run and inspected via stargraph inspect.
• Add a Fathom rule pack — wire deterministic governance onto the first graph, see the routing rule fire in the audit log, validate the pack with the Fathom plugin tooling.

Intermediate — real nodes

• Agent with retrieval — RetrievalNode fans out across a LanceDB vector store, fuses hits, and feeds a DSPyNode summariser through the force-loud DSPy adapter.
• Classical ML in a graph — MLNode running an ONNX classifier inside an otherwise-LLM graph, with the sklearn safe-load gate explained.

Advanced — production wiring

• HITL graph — InterruptNode pauses a run, stargraph respond resumes it, and the whole thing survives a cold-restart of the engine.
• Serve and replay — boot the FastAPI app with stargraph serve, enqueue a run over HTTP, and replay it byte-identically with stargraph replay.

Prerequisites for every tutorial

• Python 3.13+
• uv add stargraph (or pip install 'stargraph[stores,ml]' for the retrieval and ML lessons)
• A scratch directory to hold per-run state under ./.stargraph/

Reference docs nearby

Each tutorial links into Concepts, Reference / nodes, and the Engine and Serve sections. Read them once you understand the mechanics; the tutorials cover only what each lesson needs.