Use case
Multi-agent orchestration
Coordinate many agents as one deterministic graph — supervisors, swarms, and hand-offs that run in parallel and stay checkpointable.
The problem
- Role-based frameworks like CrewAI and AutoGen delegate between agents, but the control flow is implicit and hard to reason about or replay.
- Fan-out across many agents on an interpreted runtime spends most of its time in scheduling and message-passing overhead, not model calls.
- When one agent in a swarm fails, the whole run often has to restart from the top.
How Agentmatic solves it
Graphs, not delegation
Model a supervisor or swarm as a StateGraph. Every hand-off is an explicit edge, so the topology is deterministic, parallel, and replayable — the same API you'd use in LangGraph.
Parallel within a superstep
The Pregel runtime executes every eligible node in a step at once over lock-free SPSC channels, so wide fan-outs are handled by the Rust scheduler instead of Python's event loop.
Failure is contained
Circuit breakers, retries and dead-letter queues wrap individual agents, so one flaky sub-agent doesn't take down the run.
Keep reading
Part of the Agentmatic use cases, from Neul Labs ↗.
Ship your next agent in minutes, not weeks.
MIT licensed. Drop-in for LangGraph. Native SDKs in 5 languages. Battle-tested resilience primitives in the box.