DBOS

DBOS is a Cambridge, Massachusetts company co-founded by Postgres creator and Turing Award winner Mike Stonebraker together with a joint team of MIT and Stanford computer scientists, growing out of a multi-year research project between the two universities. Its core product, DBOS Transact, is an open source durable execution library: developers annotate workflows and steps in ordinary Python, TypeScript, Java, or Go code, connect a Postgres database, and their program becomes durable. Every step is checkpointed in Postgres, so if a process crashes or restarts, workflows resume from the last completed step instead of starting over, with exactly once semantics. There is no separate orchestration server to run, which is the company's core contrast with Temporal, and the design also positions it against Airflow for pipelines and against Celery or BullMQ for queues, since DBOS queues are durable, Postgres backed, and integrate with workflows.

Around the library sits DBOS Conductor, a workflow ops control center that visualizes workflow state and history, monitors queue metrics in real time, automatically resumes failed executions, and lets teams cancel, resume, fork, or replay workflows from a dashboard. Conductor supports custom alerts with Slack and PagerDuty integration, integrates with Redshift and Databricks Lakebase for observability data, and can be used from a coding agent for monitoring and debugging. It runs as a hosted service or self hosted on a team's own infrastructure, including air gapped environments, and never sees or stores customer workflow data. DBOS reports SOC 2, HIPAA, and GDPR compliance with single sign-on and SAML support, and recent releases added role based access control and an OpenMetrics endpoint, alongside a Google ADK plugin for agent builders. DBOS Cloud is a serverless execution platform for hosting Transact applications.

DBOS sits at the reliability layer of the agent stack rather than building agents itself. It is a strong fit for teams that want fault tolerant agents, pipelines, and background jobs with minimal rearchitecting, especially teams already on Postgres, and a weaker fit for teams that do not want Postgres in the stack or need languages the library does not yet support, a tradeoff the company itself documents.