Glossary
Approval loop
The human-in-the-loop pattern where every agent send queues for explicit human approval before going out.
The approval loop is the human-in-the-loop pattern Helix uses for every agent send. The agent drafts. The draft lands in a queue. The human approves, edits, or skips. The agent learns from the edits over time. No outbound action ships without a human nod, by default.
The default approval policy is "always" — every send waits. Per-identity, you can lower the bar (e.g. autonomous within tight guardrails for after-hours triage), but the policy is always visible in the dashboard and every action lands in the audit log with the approving human attached.
The approval loop is what separates an agent that helps from an agent that embarrasses you. Most "AI sends email on autopilot" tools either over-promise (and burn customer trust on a bad send) or under-promise (and end up as a draft generator). The approval loop is the third path: the agent does the typing, the human owns the send.
Related terms
In product