conductor

Why we included this project

Conductor started at Netflix as a way to keep microservice orchestration reliable under heavy load, and the same durability now applies to AI agent pipelines. You get a server with a UI, define workflows as code or JSON, and write workers in whatever language you like; the engine tracks state, enforces timeouts, and lets a long-running agent loop pause and resume without losing its place. It includes native LLM providers, MCP tool calling, and human-in-the-loop approval, so autonomous agents can discover tools, reason, and act until the task is done. Production teams at companies like Tesla and LinkedIn have run it for years, which is a decent sign that the failure-handling story holds up in practice.