// the find
conductor-oss/conductor
Conductor is an event driven agentic workflow engine providing durable and highly resilient execution engine for applications and AI Agents
Conductor is a durable workflow engine originally built at Netflix, now maintained by Orkes. It orchestrates microservices and AI agents using JSON-defined workflow graphs, where the engine handles all state persistence, retries, and recovery — workers are plain code with no framework constraints. Good fit for teams building long-running workflows, multi-step agent loops, or anything where 'what step was I on when the server crashed' is a real problem.
The separation between deterministic orchestration (JSON) and unconstrained worker code is the right call — you avoid the replay-safety headaches that plague code-first engines like Temporal. Battle-tested lineage from Netflix means the core execution model isn't experimental. Native LLM task types (LLM_CHAT_COMPLETE, LIST_MCP_TOOLS, CALL_MCP_TOOL) built into the engine rather than bolted on as an afterthought. Five persistence backends and six message brokers means you're not locked into a specific infrastructure stack.
JSON workflow definitions get ugly fast for anything non-trivial — the DO_WHILE example in the README already requires understanding how `$.loop['think'].output.result.done` resolves, and there's no type safety until runtime validation fails. The Redis+Elasticsearch default backend adds significant operational overhead for teams that just want Postgres. Orkes as the primary maintainer creates a commercial tension: the enterprise version exists, and some features will inevitably land there first. Ruby and Rust SDKs are 'incubating', which usually means they're incomplete and you're on your own if they break.