I’m designing a procurement automation that takes 3-4 weeks from supplier bids to final approval. My challenge comes when agents handling vendor comparisons get stuck waiting for legal review, then lose contextual awareness of previous steps. I need these AI roles to maintain situational awareness even with days between steps. How are others handling state persistence in workflows with unpredictable human delays?
Latenode’s dynamic memory sharing solved this for our contract workflows. Each agent maintains context through JSON state objects that get auto-updated across delays. Set retention policies to keep data fresh between steps. No more manual handoff docs needed.
We built a middleware cache layer using Redis to store workflow state, but maintenance became cumbersome. Recently switched to native snapshotting features in our automation tool - way more reliable for cross-agent context. Key is ensuring your platform has version-controlled state management baked in.
Implement checkpointing with expiration timestamps. At each workflow phase completion, persist these to a shared datastore accessible to all agents. Use hash comparisons during resumption to detect stale data. Crucial to balance persistence frequency against storage costs - weekly workflows can generate GBs of state data if not optimized.
just use temp databases that auto-purge after X days. but make shure agents can acces same tables. had sucess with sqlite for smaler flows