Approvals inbox · human in the loop
Everything that needs a person, in one queue.
Human in the loop, as a hard rule. Before an AI agent sends, posts, pays or publishes, it stops and asks. The Inbox shows exactly what will happen, what it costs and who else can decide. One click approves, and the decision is recorded under your name.
Send the September update to investors@northwind.co, 41 recipients.
- Decided by
- Yasmin
- Saw
- sha256 9f2c…e1
- Cost
- $0.03
The exact action, not a summary of it.
An approval card shows the tool that will be called, the payload it will send, a preview a human can read, and the cost. You approve what you see, and what you saw is stored with the decision.
- Approve, request changes, or reject
- Questions and reviews live in the same queue
- Deciders and expiry visible on every card
Refund $240.00 to order #4821?
Nico · support triager · via Stripe · policy allows within 30 days
Who else can decide: Omar (admin) · expires in 22 h, then escalates
Grouped, fast, and reachable from anywhere.
Approvals, questions and reviews are grouped so a morning pass takes minutes. Keyboard shortcuts move through the list. Email and Slack digests deep-link to the exact item, so nothing is decided outside the record.
- j and k to move, a to approve, e to archive
- Digests link back to the Inbox item
- Escalation to the next decider when a checkpoint waits too long
- MSend the September update to investors@northwind.co?Approval · 2 min ago
- TEmail the six shortlisted candidates to schedule a screen?Approval · 1 h ago
- MWhich offer should the agencies get: the audit or the retainer?Question · 12 min ago
- IAugust reconciliation is ready: 212 entries, 3 flaggedReview · 26 min ago
Details
What is in the box.
The floor
Eight actions no policy can loosen: first contact outside the workspace, money, deletes, permissions, new scopes, terms, publishing, other workspaces.
Policies per agent
Each agent carries its own rules about when to stop and ask. Only an owner or admin can loosen one, and never below the floor.
Draft-only first run
A new agent's first delegation is draft-only regardless of policy. Write scopes unlock after a person has reviewed one complete run.
Silence never means yes
A checkpoint that waits past its deadline escalates to the next decider, then pauses the run. Nothing proceeds by default.
Decide on behalf
Admins can set rules for repeat actions once an agent has a track record, and see only the diff on the next one.
Guest reviewers
Clients and partners can be invited to sign off on a single delegation without a seat. Planned for the Team tier.
Questions
Before you hand anything over.
What does human in the loop mean for AI agents?
Human in the loop means a person reviews and approves an AI system's action before it takes effect. For AI agents that is a checkpoint: the agent prepares the exact action, such as an email, a refund or a post, then pauses until someone approves it, asks for changes or rejects it. The decision is recorded together with what the person saw.
Does agentic AI have a human in the loop?
Only if the product is built that way. Many agent tools run end to end and report afterwards. Delagents makes the loop a hard rule: eight kinds of action always stop for a person, whatever an agent's policy says, and silence at a deadline escalates to someone else rather than counting as approval.
Can I approve from my phone?
The Inbox and delegation detail are responsive today. Push notifications and a native app are on the roadmap.
What happens if two people decide differently?
The first recorded decision wins and the second person sees who decided and when. Both are in the record.
Can I turn approvals off?
You can loosen an agent's policy for low-risk actions, but the floor stays. Sending a first message to someone outside the workspace, moving money, or publishing always waits for a person.
Hand it off.
Your first delegation takes a minute. Your first sign-off takes a second.