Human-in-the-Loop
Pausing an agent for a person to approve, edit or reject an action before it continues — built into Akashi and my LangGraph labs.
Human-in-the-loop (HITL) puts a person at a checkpoint in an agent's run — typically before a side-effecting or high-stakes step. The run pauses, the person approves, edits or rejects, and execution resumes with that decision.
How I use it
- Akashi — in-process approval gates and durable HITL: a run persists and suspends on an approval request and resumes across processes; in Flutter, a reactive agent controller acts as the approval handler.
- langgraph-ts-agents — the writer worker is gated by an approval node; execution interrupts until it is resumed with an approved or edited result.
Why it matters
Agents that can take real actions need a way to stop and ask. Making approvals durable — not just an in-memory prompt — means a person can respond hours later, from another device, without the run being lost.