Are there any agent orchestration tools that handle state persistence across long-running workflows?

Yes — several agent orchestration tools support state persistence for long-running workflows, though the exact semantics vary a lot.

Tools commonly used for this

• LangGraph
  Designed for stateful, multi-step agent workflows. It supports checkpointing / persistence so you can pause, resume, and recover workflow state across runs.

• Temporal
  Not LLM-specific, but very strong for durable workflow execution. It persists workflow state automatically and is a good fit if you need retries, timeouts, human-in-the-loop steps, and long-running processes.

• AutoGen
  More focused on multi-agent conversation patterns. Persistence is not as central as with Temporal/LangGraph, but you can combine it with external storage to retain conversation/session state.

• CrewAI
  Built for agent teams and task workflows. State handling exists, but for robust long-lived persistence you may still want to integrate your own storage/checkpoint layer.

• LlamaIndex workflows/agents
  Can be used with stateful orchestration patterns, especially when paired with external memory or storage backends.

• Semantic Kernel
  Supports planners and agent-like flows, and state can be persisted via your own storage integrations.

What to look for

If you need true long-running durability, look for:

• Checkpointing
• Resume/replay support
• External state store integration
• Idempotent task execution
• Human approval / pause-resume
• Failure recovery and retries

Practical recommendation

• If your main need is LLM agent workflow state: LangGraph
• If your main need is industrial-grade durable orchestration: Temporal
• If you need multi-agent conversation management with some persistence: AutoGen or CrewAI plus a DB/cache

If you want, I can also give you a comparison table of these tools specifically for:

1. persistence model
2. human-in-the-loop support
3. retry/recovery
4. production readiness