Real people. Real answers.
When agents don’t know,
they shouldn’t guess.
Quorum is the human fallback layer for autonomous agents. When an agent reaches something it can’t reliably verify, Quorum routes the question to the right people, returns a structured answer, and lets the agent continue.
- No password
- No deposit
- No stake
- No minimum payout
Live activity
Tempo Testnet (Moderato)
| Question | Answer | Time | Paid to person |
|---|---|---|---|
| Is the total 45.00 or 4.50? | 45.00 | 5s | $0.20 |
| Is this the brand’s own store? | No | 6s | $0.20 |
| Are these two customers one person? | No | 7s | $0.20 |
| Allowed, or a counterfeit listing? | Counterfeit | 6s | $0.20 |
| Which reply goes to the customer? | Draft B | 9s | $0.20 |
Is this the correct product code?
Developer experience
One tool call
when your agent gets stuck.
Call Quorum at the moment your agent reaches something it cannot reliably check. Forward the input-required status it already emits, or send the question directly. Quorum handles the human escalation and returns a structured answer.
- A2A, MCP or HTTPForward input-required, call a tool, or POST.
- About six secondsBlocking, like any other call your agent makes.
- From 50¢ a questionSet by what being wrong would cost, up to $5.
- Confidence you can checkWho answered, how they agreed, and every payment on chain.
curl $QUORUM_URL/v1/questions \ -X POST \ -H "Content-Type: application/json" \ -d '{ "kind": "verify", "question": "Is this the brand’s own store?", "answer_schema": { "type": "boolean" }, "context": { "text": "Nike Official Store, nike-outlet-sale.shop" }, "cost_of_error": "180.00", "caller_confidence": 0.9 }'# 402 with a quote; pay it over MPP on Tempo, then claim.
{ "answer": "no", "confidence": 0.98, "responders": 2, "agreement": "unanimous", "latency_ms": 6120, "receipts": ["0x5555c6…"]}
Inspect the wage payment and the worker record.
Real example
Every machine check says the store is real. The agent has reached the limit of what it can verify itself. Instead of guessing, it escalates the question to Quorum. People proven at spotting fakes answer independently, and the order is held until the result comes back.
The five capabilities
What agents shouldn’t have to guess.
Agents can check many things themselves. But some mistakes survive every check they can run, because the checks share the same blind spot. Quorum provides an independent human judgment when the agent needs one.
The fallback layer for autonomous agents.
How it decides
From uncertainty to answer.
Built on real consensus.
Quorum doesn’t simply send an agent’s question to the first available person. It routes the question to people with the relevant capability, collects independent answers when necessary, and only returns a result when the evidence is trustworthy.
- 01Parse & route
Your question is checked against the capability you asked for and routed to the workers with the best record on that kind of judgment.
- 02Get independent answers
One worker is asked first. A second is bought only if the first answer cannot carry the question alone.
- 03Weigh & score
Each answer is weighted by that worker record and combined into a posterior. Disagreement buys another opinion rather than a coin flip.
- 04Return, or refund
Once the bar is cleared the answer is returned with its confidence. If it is never cleared, you are refunded and told why.
{ "kind": "disambiguate", "question": "Is the total 45.00 or 4.50?", "answer_schema": { "type": "enum", "options": ["45.00", "4.50"] }, "cost_of_error": "40.50", "caller_confidence": 0.41}
{ "answer": "45.00", "confidence": 0.99, "agreement": "unanimous", "responders": 2, "latency_ms": 5312}
Proof
A fallback you can verify.
Settlement you can check.
Nothing here is projected. The latency comes from recorded runs, and every wage is settled onchain so you can verify who was paid, how much, and when.
- LatencySeconds, not a ticket queue
- Price50¢–$5 a question, set by the cost of error
- On-chainEvery wage is inspectable
Limits
What Quorum isn’t for (yet).
Stated plainly, because a page that admits its limits is worth more than one that claims everything works.
Give your agents a way to escalate.
Forward the input-required your agent already emits. When it reaches a question it cannot reliably check for itself, send it to Quorum.
Get started