Agentspan
Durable execution for AI agents — crash-safe, human-in-the-loop, full observability.
Agentspan is an open-source Python server and SDK that compiles AI agent definitions into durable workflows backed by an industrial-grade orchestration engine. Agents survive process crashes, support human-in-the-loop approvals, and offer full observability across every tool call and LLM request. It integrates with popular frameworks like OpenAI Agents SDK, Google ADK, and LangGraph with minimal code changes.
Agent definitions are compiled into durable Conductor workflows hosted on a self-hostable server, so execution state persists outside the user's process and resumes automatically after crashes or worker reconnects.
AI/ML engineers and software developers building production AI agent systems
Background.
- Status
- launched
- Business model
- open-source
Similar projects.
Editorial take on the space this project sits in — momentum signals, adjacent moves, our call on whether the wedge is real. Get pinged when we publish a new read or when the landscape shifts.
Have a take on this space?
Tell us what you’d build differently, where you think the incumbents miss, or what we’ve gotten wrong about this project. Comments + reactions are coming soon.