Use case
Long-running durable workflows
Agent runs that take minutes or hours shouldn't die on a deploy. Agentmatic checkpoints every superstep so work is never lost.
In one line: Checkpointing ships with five backends (Memory, SQLite, Postgres, Redis, S3) in the open-source core — see the checkpointing page for the durability model.
The problem
- Long agent runs are fragile: a crash, redeploy, or timeout throws away everything computed so far.
- Rebuilding progress means re-paying for every LLM call that already succeeded.
- Durable state is often bolted on with ad-hoc databases and brittle glue.
How Agentmatic solves it
Every superstep is a checkpoint
Because the Pregel runtime treats each superstep as a checkpoint boundary, persisting and resuming state is a first-class, cheap operation — not an afterthought.
Pluggable durability
Choose the checkpointer that fits: Memory for tests, SQLite for a single box, Postgres or Redis for shared state, S3 for large or archival runs.
Resume anywhere
Restart a crashed run from its last checkpoint, or rewind to any earlier superstep with time travel to retry a branch.
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.