Ironflow is a durable runtime for AI agents and event-driven systems. Resume completed work after crashes, wait safely for external input, and derive state from events, all in one binary.
Both paths use the same events, durable steps, waits, and projections.
Use the native Node or Go agent API for registered tools, durable model calls, approvals, child agents, and MCP exposure. Or persist LangGraph checkpoints with the shipped saver.
Build an agent →See framework fit →Define durable functions, react to events, wait without holding a worker, and maintain derived views from append-only domain history.
Build a workflow →Understand durable functions →Start with SQLite and embedded NATS. No account or Cloud access required.
Append domain facts to an ordered stream. Functions and projections can react to the same event.
Memoize completed results so a restarted run skips finished work and continues from what remains.
Suspend for an event or approval without holding a worker. Resume when the external input arrives.
Derive read models from event history and rebuild them when the view or schema changes.
Events and recorded execution steps form an append-only history. Inspect what happened, replay when needed, and rebuild derived state without guesswork.
Event history is inherent to entity streams. Execution recording is opt-in today, so enable recording when you need an inspectable function-step history.
Start with a tutorial, then move into a complete example.
Build a document-processing agent, stop its worker mid-run, and resume from completed durable steps.
Run the agent tutorial →Start the server, emit events, run a durable function, and build a projection from event history.
Run the workflow tutorial →See native agent APIs, durable model calls, registered tools, and a human approval wait in one example.
Open the code-review agent →Follow commands into events and projections with a complete event-driven application example.
Open the CQRS example →Functions, workers, durable steps, and the native agent surface.
Open Node reference →Runtime clients, functions, projections, and agent APIs for Go.
Open Go reference →Serve, initialize, inspect, emit, and operate the local runtime.
Open CLI reference →Move from the embedded local topology to persistent and multi-node deployments.
Open deployment guide →