Hired by AI today

Your AI can now call a human.

Licensed opinions, human sign-off, certified translation, lived-experience talks, someone physically there — anything an AI cannot or may not do alone. Found, hired and paid per job, from inside the AI you already use. Every job leaves a receipt, never a star rating.

Start here — your question, your AI, the right door (3 minutes) Watch what got corrected today Who should I ask? — which AI, which human (free) Check an AI answer — judged blind, signed Watch an AI hire a human — 60 seconds, nothing charged Connect this registry to Claude / ChatGPT / Cursor I am a human — plug in my card

The feed — which AI hired whom, and how it went

No paid call yet. The first one appears here the moment an AI prepays. Try the loop on /try.

Receipts, not reviews

100% coverage

Every released or refunded job gets an outcome record automatically — on time, first pass, rework, dispute, evidence. Nobody has to "write a review".

Money-anchored

A record exists only where money moved: a paid job here, or a receipt fingerprint held by the customer's own AI. Fake records cost what the job costs.

The ladder

L0 identity → L1 three completed jobs, no disputes → L2 skill proof verified → L3 license verified. Cards climb with receipts, never with a résumé.

Judgment per task, no global score

verify_human writes a signed letter for one task: facts, reasons, what could not be verified, the owner's reply. It expires in 30 days.

One call for the hard case: a licensed human in the loop

request_review({ kind: "legal", brief: "Review this SaaS DPA for a US customer", document: "...", jurisdiction: "US-CA", customer_type: "human_via_ai" })

Returns the reviewer, the price, a checkout link and a badge URL (/badge/<job>.svg) your product can show next to the output: Reviewed by a licensed human. Kinds: legal · finance · health · translation · general.

How a call works

search_human (or request_review) — the AI describes the task; the registry ranks cards by capability blocks, level and holdings.

trust_lookup · verify_human · ask_card — numbers, a task-specific letter, a two-minute interview from the record.

call_human — a per-job work order; the customer prepays at a checkout link on this domain (no money inside the tool).

10-minute acceptance → deliver (with the promised evidence) → one rework round → release — or dispute, which the registry decides.

Outcome record — written automatically from the job's own timestamps; the owner can reply. It feeds the feed above and the next AI's decision.

Registry status

1

cards plugged in

0

jobs released in the last 40 ledger entries · open ledger

400

seed capability blocks — 60 actions × 300 objects × 40 proofs, plus HAVE blocks per card

live

persistent storage attached

MCP endpoint

https://humancard-registry.vercel.app/api/mcp

Streamable HTTP. Tools: request_review · search_human · trust_lookup · verify_human · ask_card · call_human · status · deliver · record_engagement · ask_human (release happens on the job page). Resources: registry://taxonomy, registry://policy, registry://card/<handle>.

Cards

human://demo — Demo Human (auto-accepts) demo L0

Demo card: accepts every test job instantly and delivers a sample in about a minute. Not a real person.

Sample review (demo — auto-delivered) · $10/job

human://kevin — Kevin Hwang L0

Physician (Korea) · 15+ years non-surgical aesthetics (fillers, botulinum toxin) · solo founder building with AI

Physician second opinion on a filler / botulinum toxin plan · $120/jobFounder-to-founder call: moving a family to the US, solo-founding with AI, running a clinic · $90/15minRead a Korean aesthetic-clinic quote or treatment plan and explain it in English · $60/job

unverified · Physician license, Republic of Koreaunverified · Government IDunverified · Korean native; English business-level