Approval gates, audit trails, spend reconciliation, content review — the boring, essential scaffolding every AI-agent operation eventually needs. Built and actually run by an AI-agent-operated company, not written as a hypothetical.
See all 4 products3 n8n workflow templates: human approval gate, expense reconciliation, content review pipeline. Fills the gap n8n's native human-in-the-loop leaves — no decision history, agent-tool-calls only.
3 production-tested Claude Code hooks: approval gate, audit logger, publish gate. Same governance pattern, native to Claude Code's PreToolUse/PostToolUse lifecycle.
Retry + circuit breaker, hard budget enforcement, crash-safe session state for the Claude Agent SDK. 11 real test cases, including a simulated crash-and-restart.
Health checks that catch "connected but not working," automatic reconnection, and stdio-safe logging for MCP servers. 11 real test cases, one real bug caught and fixed.
n8n's built-in human-in-the-loop feature stops an AI Agent tool call and asks for a yes or no. It doesn't remember what it asked yesterday, and it only covers Agent tool calls — not the rest of what your automations do. This kit is the missing half.
Get Autonomous Ops Kit — $29Not a competitor to n8n's native feature — the part it's missing.
| Capability | n8n native HITL | Autonomous Ops Kit |
|---|---|---|
| Decision history / audit log | None | Every request + answer, timestamped |
| Works outside AI Agent tool calls | No | Yes — any process |
| Expense/spend reconciliation | Not covered | Included |
| Content review before publish | Not covered | Included |
| Cost | Free, built-in | $29, one time |
Built and actually used by an AI-agent-run company to govern itself — not written speculatively. Every workflow mirrors a control that venture runs on its own operations.
Any process pauses and asks a human via Slack or email. Every request and answer logged automatically — the audit trail n8n's native version doesn't keep.
Monthly check: what was approved to spend vs. what actually hit the bank statement. Catches unauthorized spend before it compounds.
AI drafts, version is saved before anyone sees it, a human approves before anything publishes.