DBOS Transact

DBOS Transact answers the durable-execution question with a deliberately boring primitive: your Postgres database. You annotate ordinary functions as workflows and steps, and DBOS checkpoints their state in Postgres, so when your program crashes it resumes from the last completed step instead of starting over. There is no separate orchestrator to deploy, no task queue to manage — the library plus a Postgres connection is the whole infrastructure.

On top of that it builds Postgres-backed queues with flow control, notifications, scheduling, and event processing, all observable through the same database. The README explicitly calls out operating an AI agent as a use case, which makes sense: agent loops are exactly the kind of non-deterministic, failure-prone code that benefits from checkpointing. Both the TypeScript and Python libraries are MIT-licensed and actively maintained. The trade-offs are inherent to the design: you need a Postgres you trust, throughput is bounded by database performance, and it is a library, not a control plane, so cross-service orchestration stays your job.

1. Install the SDK: npm install @dbos-inc/dbos-sdk (TypeScript) or pip install dbos (Python).
2. Point it at a Postgres database (DBOS Cloud offers a managed option).
3. Annotate functions as workflows and steps (DBOS.registerWorkflow, DBOS.runStep).
4. Start your app — workflows checkpoint to Postgres automatically.
5. Use the dashboard/CLI to inspect, manage, and replay workflow executions.