shuang

Pi-graph authors graph JSON, validates structure, runs nodes, resumes from checkpoints, and visualizes execution. It supports isolated, thread, and shared context modes, conditional routing, fan-out with barriers, structured outputs between nodes, and human approval gates for review loops.

Developers and researchers who need specialization, parallelism, independent review, or persistent role memory should use pi-graph. It fits teams coordinating researcher, writer, and reviewer passes with explicit state. Simpler tasks should stay on a single loop, per the manifest guidance.

Author the graph JSON, validate it, then run, resume, or visualize the graph. Use checkpoints for durable recovery. The manifest names these operations explicitly but does not provide install commands, environment setup, or runtime prerequisites.

Licensing and cost are not stated in the provided manifest. No repository license, pricing tier, hosted service, or dependency list is included, so confirm terms directly from the project source before adoption for production use.

No prerequisites are documented beyond needing a task that genuinely benefits from specialization, parallelism, independent review, or persistent role memory. Keep scope narrow. The manifest warns to prefer a single loop when those needs are absent.