For AI agents & human-in-the-loop pipelines

Human in the loop,
with the humans included.

Every HITL framework gives you the pause button — the approval gate where your agent stops and waits. GetABrain gives you the person who answers it. One API call reaches a real, quality-scored human and returns structured JSON in minutes.

Get an API key — $5 freeRead the docs

No card required · from $0.05 per answer · 16 response types

What is human in the loop?

Human in the loop (HITL) is any workflow where an automated system pauses to let a person review, approve, correct, or answer before it continues. It's how you keep judgment, accountability, and a safety valve in a process that's otherwise running on its own.

In AI agents it usually shows up as an approval gate: the agent reaches a risky, ambiguous, or high-stakes step — a refund, a publish, a spend, a legal call — and stops to wait for a human decision instead of guessing. It's the antidote to a model confidently grading its own homework.

But the pattern only defines the pause. The frameworks below hold your agent's state open while it waits — they don't hand you a person to actually answer. That's the gap GetABrain fills.

Your framework has the pause button. We're the people who answer it.

Keep the HITL mechanism you already use. Drop a single GetABrain call inside its approval step, and the human it's waiting for is on tap — no recruiting, no Slack channel of reviewers, no waiting on your own team.

Works inside·LangGraph interrupts·AWS Bedrock Agents·Cloudflare Agents·Microsoft Agent Framework·your own approval node

One call in. Structured judgment out.

Send the decision your agent is stuck on. Get real human answers back as JSON your code can act on — with the reasoning attached.

human-in-the-loop approval step
POST https://www.getabrain.ai/api/v1/requestor/queries
X-API-Key: gab_live_...

{
  "type": "yes_no",
  "title": "Approve this refund?",
  "content_data": {
    "question": "Customer asks for a refund on order #8842 after 40 days. Policy is 30 days but they cite a defect. Approve?"
  },
  "required_responses": 3,
  "bid_amount_cents": 25
}

// minutes later — delivered to your webhook, or on poll:
{
  "status": "completed",
  "responses": [
    { "answer": "yes", "explanation": "Defect overrides the window; goodwill keeps the customer." },
    { "answer": "yes", "explanation": "Documented defect — refund is the right call." },
    { "answer": "no",  "explanation": "Past policy; offer store credit instead." }
  ]
}

How the loop closes

1

Your agent hits a gate

At any approval, ambiguity, or judgment step, your code POSTs the question to GetABrain instead of guessing — or halting on your own team.

2

Real humans answer

Quality-scored people who match your requirements respond in minutes. Require a minimum tier for sensitive calls; ask for as many independent answers as you want.

3

You get structured JSON

Each answer arrives as clean JSON with the reasoning attached — polled or pushed to a signed webhook — and your agent continues, now with human judgment in hand.

16
response types
minutes
typical turnaround
$0.05
per answer, and up
$5
free trial credit

Human-in-the-loop questions, answered

Human in the loop (HITL) is any workflow where a machine pauses to let a person review, approve, correct, or answer before it continues. In AI agents it usually means an approval gate: the agent stops at a risky or ambiguous step and waits for a human decision. The pattern gives you oversight and a safety valve — but on its own it only defines the pause. You still have to supply the human who fills it.
Frameworks like LangGraph, AWS Bedrock Agents, Cloudflare Agents, and the Microsoft Agent Framework give you the mechanism to pause an agent and wait for input — the interrupt, the approval node, the state that survives the wait. What they do not give you is a person to actually answer. GetABrain is the supply side: one API call from inside your existing HITL step reaches a real, quality-scored human and returns a structured answer. Use them together — their pause button, our people.
Most simple queries (yes/no, multiple choice, a rating, a short judgment) come back in minutes, not days. You set the number of responses you need, a per-response bid, and an expiration window; higher bids and simpler questions clear faster. You can poll the query or receive a signed webhook the moment each answer lands.
Sixteen response types, including yes/no, multiple choice, rating scales, ranking, free-text judgment, comparison, categorization, sentiment, and voice, video, or photo capture. Anything where you want real human judgment rather than another model grading itself.
Every worker carries a quality score and tier, verifies email and phone at signup, and passes identity verification before any payout. Responses below per-type time floors are rejected, and low-rated workers are suspended automatically. You can require a minimum quality tier per query so sensitive approvals only reach your best-rated people.
Prepaid, pay-per-answer, from $0.05 per response with a 15% platform fee included — no subscription or minimums. New accounts start with $5 of free trial credit and no card required, so your agent can make its first real HITL call before you spend anything.

Put a real human in your loop today.

Sign up, grab an API key, and make your first human-in-the-loop call with $5 of free credit. No card, no sales call.

Get your API key