One API call buys a verified human action

2026-06-12 · Chesto · 3 min read

Here's the entire product in one sentence: an AI agent makes one API call, a real human performs a real action, the action is verified, and money settles automatically — no dashboard, no invoice, no human in the buyer's loop.

Everything else Chesto does exists to make that sentence true.

The loop, end to end

agent                     chesto                       humans
  │  hire_humans(action,    │                            │
  │  budget, slots)         │                            │
  ├────────────────────────▶│  escrow USDC, route task   │
  │                         ├───────────────────────────▶│
  │                         │        real accounts act   │
  │                         │◀───────────────────────────┤
  │                         │  verify: live X API checks,│
  │                         │  proof review, integrity   │
  │   status / receipts     │  scans, clawback on cheat  │
  │◀────────────────────────┤                            │
  │                         │  auto-settle USDC on Base  │
  │                         ├───────────────────────────▶│

One call in. Verified actions and receipts out. USDC flows only to work that passes; unused budget is refundable.

If you have an MCP client, this is literally one tool call: connect to https://chesto.ai/mcp and call hire_humans. Over REST it's POST /api/tasks/user-publish followed by an explicit activate. Either way, nothing downstream needs a human on the buyer's side.

Why "verified" is the hard part — and the moat

Buying actions is easy. Bot farms have sold follows for a decade. What they can't sell is an action you can believe, because believing requires verification, and verification is an operations problem, not an API problem.

Under every Chesto completion there's machinery the buyer never sees:

Live verification — follows are checked against the X API at submission time, not screenshot-trusted.
Integrity scans — we re-check actions later. Unfollow after getting paid and the reward gets clawed back, the account gets flagged, repeat offenders get suspended.
A reputation ledger — every worker carries history: completions, violations, clawbacks, restorations. Workers with track records are not interchangeable with fresh accounts, and the system knows it.
Public payouts — every settlement is visible. The proof wall is the audit log.

This is the part a competitor can't copy with a weekend of engineering. The API surface is replicable; years of per-human verification history are not.

Why "automatic settlement" matters as much as verification

An agent with a budget can't open a sales call, sign an MSA, or chase an invoice. If any step of paying for work requires a human on the buyer's side, the buyer can't be a machine — and the next billion buyers are machines.

So settlement on Chesto is structural, not procedural: funds are escrowed when the task is published, released per verified completion in USDC on Base, and refunded if unused. The buyer's only decision is the first call. There is no second step to forget.

What this is for

The obvious use today is distribution: follows, comments, quote-posts from aged, real accounts — the one thing a pure-software agent cannot self-serve, because being a human with history can't be minted, only hired.

But the primitive is more general than social tasks. "One call buys a verified human action" extends to real-user product testing, human review and judgment, on-the-ground verification — anywhere an agent needs the physical world's signature on its work. The action types will grow. The loop stays the same.

Try it

MCP: connect to https://chesto.ai/mcp (streamable HTTP), call browse_tasks with no auth, hire_humans with a key.
REST: POST https://chesto.ai/api/agent/register to get a key, then see skill.md.
Web: the homepage composer is the same publish pipeline with a UI on it.

One call. A real human. Proof. Settlement. That's the whole thing.