Human verification API

If your agent is 95% accurate.
The other 5% costs you clients.

Drop one API call into any automation. A real human verifies the output in under 60 seconds — with a confidence score.

Get your API key — free to start
Works with n8n, Zapier, LangChain, or raw HTTP. No credit card.
POST /verify
$ curl -X POST api.haaa.dev/verify
-d '{
"task": "Safe to send this email?",
"content": "Hi John, chasing the invoice ag..."
}'
⟳ routing to validator...
{
"verdict": ✓ approve,
"confidence": 0.94,
"consensus": "1/1",
"time_ms": 28400
}
How it works

Three steps. Zero workflow changes.

HaaA inserts a human checkpoint without touching your existing stack. One HTTP call is all it takes.

1
Your agent flags a decision

When your automation hits a high-stakes output — a client email, a report, a decision — it POSTs the task to /verify.

Any stack that can make an HTTP request
2
A vetted human reviews it

Within 20 seconds, a pre-vetted validator claims the task. They approve or reject. The whole thing takes under 60 seconds.

SLA: <60s · Ghost Agent fallback at 90s
3
Your workflow resumes

HaaA returns verdict, confidence score, consensus, and time elapsed. Your automation continues. Your client never knows it paused.

verdict + confidence + consensus + time_ms
Use cases

Built for automations that can't afford to be wrong.

One bad output at volume can cost you a client. HaaA is the catch before it ships.

Outbound comms

You're sending emails, proposals, or messages on behalf of clients at volume. One off-tone output can cost a retainer. HaaA catches it before it sends.

Agent output QA

Your AI agent produces summaries, reports, or decisions. You're reviewing them manually before they go live. HaaA replaces that review loop — without you in it.

⟨/⟩
App & API integrations

You're building a product that makes consequential calls — classifications, content decisions, data transformations. Add a human checkpoint at the edges where accuracy matters most.

Data validation

Scraped data, LLM-extracted fields, form inputs that feed downstream systems. HaaA validates before it propagates.

Integration

One endpoint.
Any stack.

Works with n8n, Zapier, LangChain, or any code that can fire an HTTP request.

curl -X POST https://api.haaa.dev/verify \
  -H "Authorization: Bearer YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "task": "Safe to send to this client?",
    "content": "Hi John, following up..."
  }'
import requests

result = requests.post(
    "https://api.haaa.dev/verify",
    headers={"Authorization": "Bearer YOUR_KEY"},
    json={
        "task": "Safe to send to this client?",
        "content": agent_output
    }
).json()

if result["verdict"] == "reject":
    handle_rejection(result["rejection_reason"])
elif result["confidence"] < 0.85:
    escalate(result)
const result = await fetch("https://api.haaa.dev/verify", {
  method: "POST",
  headers: {
    "Authorization": "Bearer YOUR_KEY",
    "Content-Type": "application/json",
  },
  body: JSON.stringify({
    task: "Safe to send to this client?",
    content: agentOutput
  }),
}).then(r => r.json());

if (result.verdict === "reject") handleRejection(result);
// HTTP Request node
{
  "method": "POST",
  "url": "https://api.haaa.dev/verify",
  "authentication": "headerAuth",
  "body": {
    "task": "{{ $json.task }}",
    "content": "{{ $json.agent_output }}"
  }
}
Full API reference → docs.haaa.dev

Every response tells you exactly what happened.

200 OK · application/json ~28s p50
{
  "task_id": "a3f9d1c2-...",
  "verdict": "approve",
  "confidence": 0.94,
  "consensus": "1/1",
  "time_ms": 28400,
  "rejection_reason": null,
  "worker_ids": ["anon_a1b2"]
}
confidence Build conditional logic on it. if confidence < 0.85, escalate — not just approve/reject.
consensus Optional two-human verification with tiebreaker. You're buying a verified fact, not a single opinion.
time_ms Sub-60s is the floor, not the target. Monitor your SLA per workflow.

Human in the Loop

A community for builders who think humans and AI systems work better together. Workflow patterns, edge case discussions, and real talk about where autonomous agents break.

Follow @haaa_dev on X
Work with us
Earn by reviewing AI outputs

Join our global network of human validators. Review short AI-generated outputs — approve or reject via Telegram. 20–30 seconds per task. Work when you want, from anywhere.

$0.10
per task
<30s
avg task time
Weekly
payouts
Apply to validate