Pi Flow
A self-designing, durable, self-improving orchestration substrate: a graph of full-agent (pi) nodes an agent designs, an efficient fleet runs, and a control plane observes and improves.
Pi Flow lets you prove a workflow once on Claude Code, then run the identical DAG on a fleet of efficient, non-Claude models — no rewrite, no codegen, no drift.
A workflow is data, not a UI. A Claude Code agent owns the whole loop through the @piflow/core
SDK and the piflowctl CLI: it designs the DAG, spawns the fleet, monitors every node, and improves
the flow between runs. You steer by talking to the agents in the terminal — never by wiring nodes on
a canvas.
Start here
- Getting started — install and run your first workflow
- Quickstart — the shortest path to a live run
- Architecture — the substrate and the three levels
The three levels
| Level | What it is |
|---|---|
| L1 — the node | one agent fully described by a declarative envelope (work · sandbox · tools · hooks · contract). One headless pi. |
| L2 — COMPOSE | an agent designs the flow and emits a flat WorkflowSpec; the SDK compiles it to a DAG. |
| L3 — control plane | run · observe · intervene · learn — control nodes that live on the seams between nodes. |
The three skills
Pi Flow surfaces as three Claude Code skills:
piflow-init— create a workflow, or port one from Claude.js/ n8n / a fresh design.piflow-start— run and monitor a workflow on the pi fleet.piflow-enhance— improve a node, or the chain between nodes, between runs.