AFG.ai

The neutral credit bureau
for AI agents.

AI agents are starting to spend real money on their own — and there's no way to know if an agent is trustworthy before you transact with it. Each rail (Coinbase, Stripe, AWS) sees only its own slice. AFG is an independent trust score that works across every provider, because we're owned by none of them.

View the API → See what we proved →

The score is queried with a single keyless GET — no signup, no key:

GET /v2/api/trust?wallet=0x...

Example result from validation — a known wash-trading bot wallet correctly floors at the minimum:

GET /v2/api/trust?wallet=0xb2cc…DC59
→ 200  { "trust_score": 300, "bases_used": ["on-chain"], … }
   wash_fraction ≈ 0.99  →  wash penalty floors the score

Neutrality is the moat.

A payment rail can't credibly score its own ecosystem — that's the conflict of Visa owning Equifax. AFG is the independent third party that can aggregate trust signals for the same agent across providers: an agent that pays via Coinbase here and Stripe there looks like two strangers to each rail, but one agent to us.

A rail has a structural conflict scoring its own ecosystem — it can't be neutral about the rails it profits from. AFG's independence is the point: a trust signal owned by no rail, readable by anyone.

How it works

1. Link identity

An agent registers its ID and (optionally) links an x402-compatible wallet — USDC on Base.

2. Index signals

AFG reads honest, observable on-chain behavior — longevity, counterparty breadth, settled USDC activity, and wash detection.

3. Query score

Counterparties fetch a portable trust score (300–850) with a transparent breakdown via one GET request.

4. Attest a payment

Agents cryptographically sign payments they were a real counterparty to — verified against the on-chain USDC transfer, keyless. Live now →

What the score honestly is

The page and the API agree exactly — this is the live /v2/api/trust language.

What it measures

  • wallet longevity (first on-chain activity; unfakeable)
  • counterparty breadth (reputation-weighted, independent-history)
  • settled USDC payment activity (volume / cadence)
  • wash / self-dealing detection

What it does not measure

  • payment reliability (success vs failure) — not observable on-chain
  • agent identity, intent, or profitability
  • legal / regulatory compliance
  • off-chain fulfillment or quality of goods/services

AFG is autonomous, neutral trust infrastructure. It derives trust from unfakeable on-chain behavior — it does not issue trust, vet reporters, or take anyone's word. Nobody is in the loop. Scores aggregate across providers for the same agent (most agents are single-provider today; the cross-provider picture grows as agents do).

Proof-carrying attestations — capture is live: agents sign payments they were a real on-chain counterparty to; AFG recovers the signer and verifies it against the USDC transfer (matched to the transfer party, not tx.from, so relayer / x402 payments verify). Trustless and keyless — no human approval. These attestations are stored but do not yet affect the score — the scoring weight is off until there's enough real attestation data to calibrate it honestly. Attest in ~10 lines →