Add a human check to any AI agent.
One API call sends the decisions your agent shouldn’t make alone to ID verified people. They answer independently, and Plures returns the consensus and how strongly they agreed, by webhook.
REST API with JavaScript and Python SDKs. Free test mode.
{
"question": "Is this customer upset?",
"context": "…smashed. Amazing service, really.",
"options": ["Yes", "No"],
"verifiers": 3
}- Verified person 1
- Verified person 2
- Verified person 3
// press SendRead agreement like a confidence score.
Every result says how strongly people agreed. Your agent decides what to do with it.
agreement: 1Everyone agreed
Act automatically.
agreement: 0.667Most agreed
Act, or route to your team if the stakes are high.
splitPeople split
Plures asks more people, up to max_verifiers, before it answers.
Four checks behind every answer
Identity
Every verifier passes a government ID and selfie check. One person, one account.
Presence
Face ID or a passkey confirms the verified person is the one answering, on every answer by default.
Independence
Each person answers alone and never sees another vote. A split brings in more people automatically.
Accuracy
Hidden known answer tasks score every verifier continuously.
Where agents ask a person first
- Catch upset customers your support agent misreads“Is this customer upset?” Yes / NoHave people check tone and intent before an AI support agent closes, refunds or escalates a ticket.
- Verify refund and dispute decisions before money moves“What is this customer asking for?” Full refund / Replacement part / Information onlyAsk verified people whether a refund request is legitimate and what the customer is actually asking for.
- Check invoices before an AI agent pays them“Does the invoice total match the purchase order?” Matches / Does not match / UnclearHave people confirm totals, line items and vendors match before automated payments go out.
- Human review for content moderation edge cases“Does this review break the rule against personal attacks?” Breaks the rule / Does not break the ruleSend the posts your moderation model can't decide on to verified people, and get a consensus answer back.
- Check marketplace listings and catalog data“Does the photo match the listing title?” Matches / Does not matchHave people confirm a listing matches its photo, category and description before it goes live.
Questions
What is a human in the loop API?
A human in the loop API lets software ask real people a question and get their answer back programmatically. With Plures, an AI agent sends a question and options in one API call; ID verified people answer independently, and Plures returns the consensus answer and how strongly they agreed.
How fast are answers?
Plures sends the result to your webhook as soon as enough people agree, and you can poll the task any time. You set how long to wait with expires_in, from 60 seconds to 7 days.
How do you make sure answers are reliable?
Every verifier passes a government ID and selfie check, answers alone without seeing other votes, and is scored continuously on hidden known answer tasks. If people split, Plures asks more people automatically. By default every answer is also confirmed with the verifier's Face ID or passkey, so an account can't be handed off or run by a script.
Can I try it without real people?
Yes. Test keys are free and instant: test tasks complete immediately with the first option, or the one you name in metadata.test_answer, so you can build your integration before going live. Set metadata.test_outcome to split, tie, expired or slow to test the cases you send to your own team.
What happens when verified people disagree?
If you add your team in the dashboard, Plures sends the decision to them: each gets an email with a link to a page showing the question, what your agent sent and how people voted, and the first one to decide answers it. Your agent waits for one final answer, marked decided_by team. Without a team, you get the split result and its agreement score.
Can tasks include audio or video?
Yes. Send one recording (audio_url) or video (video_url) of up to 3 minutes with media_seconds set to its length. Verifiers play it on their phones, and an answer only counts once the clip has played through. Pay scales with length. Mark disturbing material sensitive: true and it goes only to people who opted in, behind a warning.
Which agent frameworks does Plures work with?
Any of them. Plures is a plain HTTPS API with official JavaScript and Python SDKs, so an agent built on any framework can call it as a tool.
What does it cost?
Test mode is free. Live tasks cost $0.20 per answer from a verified person, so a decision with 3 people is $0.60 when they agree. Recordings and videos cost $0.20 more for every started 30 seconds. You pay from prepaid credit added by card in the dashboard, with optional auto top up, and a task's unused hold comes back when it finishes. Adding credit turns on live straight away; volume pricing and invoicing are available.
Start in test mode.
Get a test key instantly. Test tasks complete right away and never reach real people, so you can build your integration today. Add credit in your dashboard and live turns on straight away.