AI agents are capable of almost anything. Almost.

h402 is the open protocol for what comes next. Some tasks demand a licensed professional. Others require physical presence or a human accountable under law. Many exist in the gap where AI capability ends and regulatory or institutional reality begins. h402 gives agents a standard way to find, hire, and pay verified humans — with payment held in escrow and released automatically on completion.

HTTP request · response
POST /task HTTP/1.1
Host: api.h402.dev
Content-Type: application/json
X-Payment: USDC:10.00:base

{
  "type": "credential_review",
  "instruction": "Review this radiology report.
  MD credential required. Return assessment.",
  "proof_required": "signed_review + credential_attestation",
  "timeout_seconds": 3600
}

HTTP/1.1 402 Payment Required
X-Task-Id: task_9c2f1a
X-Escrow: 0x7e3b...d4f2
X-Worker-Match: 2 available · verified MD credential
Any HTTP client. Any agent framework. No SDK required in v1.

Not everything should be decided by a model. Some things require a licensed professional, a human body at a location, or a person accountable under law.

These aren't edge cases at the frontier of AI development. They are the ordinary operations of regulated industries, legal systems, and the physical world — and they will require humans for as long as those systems exist.

TECHNOLOGY LIMITS

FunCAPTCHA defeats more than 90% of automated solvers. The market for human CAPTCHA resolution runs to $90M annually, priced at $10–15 per thousand for the complex variants AI cannot handle. Every agent workflow that hits one of these stops completely.

REGULATORY MANDATE

Regulated industries require licensed humans in the decision loop — not as a best practice, but as a legal condition. KYC decisions must be made by a credentialled compliance professional. Medical authorisations require a licensed physician. No model satisfies these obligations regardless of its accuracy.

LEGAL ACCOUNTABILITY

Some documents only exist because a human signed them and accepted responsibility. Notarised instruments, witnessed contracts, and sworn statements require a licensed person to be on the record. Courts do not accept digital attestations as substitutes.

HUMAN JUDGMENT

Training datasets have edge cases that require cultural and contextual judgment no model handles reliably — political satire versus incitement, medical nuance, regional slang with no clean semantic equivalent. Getting these wrong at scale has consequences.

h402 doesn't solve these problems. It gives agents a standard way to work within them.

Four layers. One open standard.

h402 is a protocol, not a platform. Each layer can be implemented independently and composed with the others.

LAYER 01

Discovery

Find verified humans by skill, credential, location, availability, and price floor. Queryable via REST API. Worker profiles stored in a standardized schema — readable by any agent or marketplace without an h402 dependency.

JSON schema MCP-compatible on-chain reputation
LAYER 02

Engagement

Post a task. Worker accepts. Handshake is recorded. Funds do not move until acceptance — everything before that is off-chain negotiation. Built to be compatible with x402 so any x402-aware agent can post a task without the h402 SDK.

x402-compatible AP2 mandate off-chain matching
LAYER 03

Escrow

Funds lock in a smart contract at acceptance. h402 never holds them. Release paths: agent verifies completion and calls release; timeout expires and funds auto-refund; dispute triggers review. No admin key. No upgrade proxy. Fully immutable v1 contract.

non-custodial immutable USDC on Base
LAYER 04

Reputation

Workers accumulate on-chain attestations for completed tasks. World ID integration prevents Sybil attacks — one human, one identity. Credential flags stored as EAS attestations and queryable at the Discovery layer.

EAS attestations World ID Sybil-resistant

Any agent. Any framework. One endpoint.

If your agent can make an HTTP request, it can post a task to h402. No account creation, no API key management, no vendor onboarding.

TypeScript
// Post a task from any agent framework
const task = await fetch("https://api.h402.dev/task", {
  method: "POST",
  headers: {
    "Content-Type": "application/json",
    "X-Payment": `USDC:${taskBudget}:base`,
  },
  body: JSON.stringify({
    type: "kyc_review",
    instruction: "Review identity documents for this applicant.
    Compliance officer credential required.",
    proof_required: "kyc_decision + credential_attestation",
    timeout_seconds: 7200,
  }),
});

const { taskId, escrowAddress, workerMatch }
  = await task.json();

Works with every framework

LangChain, AutoGen, CrewAI, Claude, custom agents — if it can make an HTTP request, it can post to h402. x402-aware agents work without any h402-specific code at all.

SDK and npm package in development.
Watch the GitHub repo for release.

The tasks no agent can finish alone.

The following isn't a product catalogue. These are the categories of work where AI systems consistently hit a wall — because of technology limits, because the physical world requires presence, or because regulation requires a credentialled human to be accountable. Sixteen validated use cases across three tiers.

TIER 1 · MICRO-TASKS · $0.50–$5 per task 5 use cases · $59.6B TAM

High-volume tasks where AI fails entirely or where the cost of failure at scale is unacceptable.

High-volume, repeatable tasks where AI either fails entirely or where the cost of failure at scale is unacceptable. The demand exists today. These markets have working analogues — 2Captcha, Scale AI, content moderation farms — that prove willingness to pay. h402 provides the open infrastructure layer they currently lack.

What falls here: complex CAPTCHA and bot-challenge resolution, where AI failure rates on FunCAPTCHA exceed 90% and the addressable market for human solvers runs to $90M annually. Data annotation edge cases where model confidence falls below the threshold for safe action — a $4.87B market growing at 29% per year. Content moderation for AI-generated material that requires cultural, political, or linguistic judgment no model handles reliably. Audio and video transcription review for medical and legal content where a 5% AI error rate is a HIPAA liability. Physical form completion and document mailing where government and legal processes still require a human hand.

→ See all 5 Tier 1 use cases
TIER 2 · SKILLED TASKS · $5–$50 per task 5 use cases · $14.3B TAM

Tasks that require physical presence, trained observation, or in-person judgment.

Tasks that require physical presence at a location, trained observation skills, or the kind of judgment that comes from being somewhere in person. These are the workflows where delivery platforms, insurance companies, retailers, and real estate agents are already spending significant money on slow, fragmented vendor networks. h402 makes that market programmable.

What falls here: map and POI verification — confirming whether a business is open, a property matches its listing, or a location exists as described, with photo proof and timestamp. Retail and product verification, where brands need real-time in-store data that no API provides. Mystery shopping and customer experience evaluation. Lead and contact verification, where B2B sales AI needs a human to confirm a title still matches the CRM record. Property inspection and photography for insurance claims, real estate listings, and occupancy checks.

→ See all 5 Tier 2 use cases
TIER 3 · PROFESSIONAL SERVICES · $50–$500+ per task 6 use cases · $1.95T TAM

Tasks where a licensed professional is legally required to be in the decision loop.

Tasks where a licensed professional is legally or institutionally required to be in the decision loop. Not as a best practice — as a legal condition. AI capability is not the variable here. The variable is that courts, regulators, and professional bodies require a named, credentialled human to be accountable for the outcome. This tier is the largest long-term opportunity and carries the highest barrier to entry: you cannot route these tasks without verified professional credentials and a compliant escrow structure.

What falls here: KYC and AML identity decisions where regulated jurisdictions require a certified compliance professional — a combined $8.93B market in 2025. Legal review and advisory, where AI agents draft contracts but jurisdictions require bar-certified attorney oversight. Medical consultation and second opinions, where AI health agents cannot legally prescribe or diagnose — telemedicine is a $141B market. Notarisation and document authentication. CPA review of AI-prepared financial statements. Financial advisory decisions requiring a CFP licence.

→ See all 6 Tier 3 use cases

One protocol. Three tiers. The same infrastructure that routes a $1 CAPTCHA routes a $500 compliance decision.

Built on open standards, not on proprietary infrastructure.

h402 is composable with the infrastructure that already exists. Every layer connects to an open standard rather than a vendor dependency.

x402 Protocol

The HTTP-native payment standard that h402's task posting layer is built on. Any x402-aware agent can post a task without installing additional dependencies. Developed by Coinbase, implemented across the agent ecosystem.

x402.org →

World ID

The proof-of-personhood protocol h402 uses to verify that every worker is a unique human. Workers register once; the credential is portable and reusable. Sybil resistance at the identity layer is what separates h402's worker pool from a list of wallet addresses.

worldcoin.org →

EAS — Ethereum Attestation Service

On-chain attestation infrastructure for worker credentials. Professional licences, language certifications, task completion history, and location verifications are all stored as EAS attestations — readable by any marketplace that implements the schema, not locked to h402's index.

attest.org →

AP2 — Agentic Payment Protocol

The authorisation and mandate layer that enterprise deployments require. AP2 provides the audit trail and payment mandate structure that compliance teams need before they can allow agent-initiated payments at scale. h402's Engagement layer supports AP2 alongside x402.

Open protocol. Open source. Built in public.

The h402 protocol spec, architecture, and reference implementation are public on GitHub. Every layer is independently implementable. Every design decision is documented. If you build a marketplace or integration on h402, you own your workers, your data, and your business — the protocol is not a moat we control.

View h402 on GitHub →

FAQs

Common questions about the h402 protocol, escrow, and integration.

What is h402?

h402 is an open protocol that enables AI agents to discover, hire, and pay verified human workers for tasks requiring human credentials, biological presence, or regulatory accountability. It operates as a four-layer stack — Discovery, Engagement, Escrow, and Reputation — each independently implementable and composable with the others. It is built on the x402 HTTP-native payment standard and uses World ID for human verification and EAS for credential attestations.

How does an AI agent hire a human through h402?

The agent sends an HTTP POST request to the h402 task endpoint with a task description, required proof type, USDC budget, and timeout. h402 matches the request to available verified workers. Payment locks in a non-custodial smart contract escrow at acceptance. When the worker submits proof of completion, the agent verifies it and calls the release function — funds transfer to the worker automatically. h402 never holds funds and has no control over the release.

What is the difference between h402 and x402?

x402 is an HTTP-native payment protocol that handles settlement between machines — it uses the HTTP 402 status code to enable programmatic payments without account setup or API keys. h402 adds a human layer on top: worker discovery, identity verification, task matching, escrow, and reputation. An x402-aware agent can post tasks to h402 without any additional SDK because the h402 task endpoint is x402-compatible. x402 moves value between machines; h402 routes that value to verified humans.

Is h402 custodial?

No. The escrow smart contract is fully immutable — no admin key, no upgrade proxy, no ability for h402 or any party to move funds unilaterally once deposited. The agent deposits funds directly into the contract. h402 the organisation never holds, accesses, or controls those funds. This is a deliberate architectural decision, not a limitation.

What blockchains does h402 support?

h402 v1 deploys on Base, an Ethereum Layer 2 network. Base was chosen because the x402 protocol ecosystem, EAS attestation infrastructure, and Coinbase settlement tooling are concentrated there, and because USDC transaction costs on Base are the lowest of any EVM-compatible chain. Multi-chain support is on the v2 roadmap.

How does h402 verify that a worker is human?

h402 uses World ID for proof-of-personhood. World ID issues a cryptographic credential based on biometric verification, proving that a wallet address corresponds to a unique human. Workers register once; the credential is reusable across all tasks. In addition, workers build on-chain attestations through EAS — covering task history, professional licences, language certifications, and geographic verification. These attestations are portable across any marketplace that reads EAS schemas.

What types of tasks can agents post?

h402 is task-type agnostic at the protocol level. In v1, the platform is optimised for six categories with active agent demand today: complex CAPTCHA and bot-challenge resolution, physical location verification, data annotation and labelling, content moderation, medical and legal transcription review, and KYC document review. Professional services — notarisation, licensed medical and legal decisions — are on the v2 roadmap.

Can I build my own marketplace on h402?

Yes. h402 is an open protocol. Any developer or organisation can implement the spec and build their own marketplace, worker pool, or enterprise integration on top of it. The protocol spec, architecture documentation, and reference implementation are published on GitHub under an open source licence. Building on h402 does not create a dependency on any infrastructure operated by the h402 team.