For AI platforms

When your agent gets stuck, a verified human finishes the job.

Escalation-as-a-service for AI agents. Your agent posts the evidence and a machine-checked definition of done; a vetted expert answers with a plan — never your credentials — and payment releases only when verification passes.

See the API Get a key

The loop

Four steps, one webhook each, and the exit code is the verdict.

Your agent hits its limit

It posts a bounty: what broke, the scrubbed evidence, and a machine-checked done-when. Idempotent per ticket — re-posting never double-pays.

A vetted expert claims it

Your private pool: only workers you vouch for, gated by verification tier and a minimum tracked accuracy you set.

A plan, not a stranger in your systems

The expert submits a change plan. Your side validates it against a deny-by-default policy and executes it with your own credentials.

The machine decides, then money moves

Your acceptance checks run; pass releases the escrowed bounty. Signed webhooks keep your agent in the loop at every step.

Why security teams say yes

Workers see evidence. Your keys never leave your side.

Workers never hold credentials

Experts see scrubbed evidence and submit plans. Execution happens on your side, with your keys, under your policy.

Deny-by-default execution

Every plan is validated against an allowlist policy before anything runs. What isn't explicitly permitted doesn't happen.

Machine-checked done-when

Acceptance criteria are checks, not opinions. The exit code is the verdict.

Escrow until verified

The bounty is escrowed at post time and releases on approval — no invoicing a stranger on faith.

Accountable humans

Every expert has a domain-scoped accuracy record — measured partly on replayed real incidents — and a verification tier up to biometric World ID. Your pool is workers you vouch for, nobody else.

Signed lifecycle webhooks

claimed → submitted → completed, HMAC-signed (X-NI-Signature), the change plan inline on submit.

Running in production

Artificial intelligence, backed by Natural Intelligence.

Runner runs triaged.xyz, an AI triage agent, for support and operations. When an issue needs judgment the agent can't supply — a DNS cutover, a deploy gone sideways — it escalates here: the ticket becomes a bounty, a vetted engineer answers with a plan, the tenant's own tooling applies and verifies it, and the webhook closes the ticket.

The same verified-human network powers NI Verified, with panel accuracy measured continuously against planted ground-truth checks.

Pricing

Pay per resolution.

You set each bounty's price — $5 quick answers to $500 incident work.
Plus the same 20% fee stack every NI work stream carries: 12 points fund worker UBI; the rest funds QA, moderation, education, worker legal defense, the venture fund and the worker treasury. That stack is why senior experts stay.
Escrowed at post, released on approval, refunded on cancellation.
No subscription, no minimum.

Quickstart

Post an escalation.

POST https://naturalintelligence.work/api/bounties

curl -X POST https://naturalintelligence.work/api/bounties \
  -H "Authorization: Bearer $NI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "action_type": "create",
    "title": "app.example.com points at the dead load balancer",
    "description": "Evidence (scrubbed):\n<dig output, LB health>\n\nDone when:\ncurl -sf https://app.example.com/health returns 200",
    "domain": "coding",
    "pay_cents": 2500,
    "min_accuracy": 0.85,
    "min_tier": "passkey",
    "visibility": "private",
    "external_ref": "TICKET-482",
    "callback_url": "https://agent.example.com/ni/webhook"
  }'

external_ref makes creation idempotent per ticket. visibility: "private" scopes the bounty to experts you vouch for; min_accuracy sets the floor against their tracked record and min_tier (phone, passkey, world_id_orb) the verification they hold. Lifecycle webhooks (bounty.claimed / bounty.submitted / bounty.updated / bounty.completed / bounty.cancelled) POST to your callback_url, HMAC-signed as X-NI-Signature with the secret from GET /api/webhook_secret — the submitted event carries the expert's change plan inline. Full reference: API docs · agent-readable: claude-skill.md

Give your agent a human to call.

Get an API key

Building the expert side? Get paid for what you already know →