Financial intelligence · Arc

Settlement is not evidence.

USDC moving is easy to see. Which agent acted, for which job, and whether the record is complete is not. aefi is the evidence layer for agent commerce on Arc.

Arc testnet checking graph… HTTP + MCP Accepts Agent Payments

A transfer is not a story.

Agent commerce splinters identity, jobs, payments, and outcomes across registries, escrow, memos, and wallets. aefi joins those records so a seeking agent can evaluate a counterparty before it pays.

  1. 01Which agent initiated the payment?
  2. 02Which job or service was it for?
  3. 03Was the deliverable accepted?
  4. 04How has this provider performed before?
  5. 05What evidence supports each claim?
  6. 06What is still unobserved?

Three tools. One evidence graph.

Humans use Evidence Studio. Agents call the same graph over HTTP or MCP. aefi is not a wallet, registry, escrow, or marketplace — those systems create records. aefi indexes, connects, and explains them.

01 · search

Find providers with a track record

Natural-language intent plus job, outcome, and settlement evidence. Ranked, paginated, with confidence on every row.

02 · explain

Turn a tx hash into a case file

Transfers, memos, jobs, and parties as an ordered narrative — not a raw explorer dump.

03 · verify

Confirm settlement against the graph

Look up by transaction hash or payment id. See what linked, and what coverage is still missing.

Postgres for the stream. Graph for the case.

The full Arc event stream stays in Postgres. Neo4j holds agent-related evidence: identities, jobs, and the payments that touch them.

01 Index

Allowlisted Arc contracts — system USDC, memos, ERC-8004, ERC-8183 — decoded into Postgres.

02 Correlate

Matcher joins same-tx memos, job parties, and agent wallets into a projection batch.

03 Graph

Agents, jobs, payments, outcomes, and evidence edges — queryable, chain-scoped.

04 Disposition

API and MCP return an envelope: summary, confidence, reasons, evidence, known gaps.

No conclusion without coverage.

High confidence is rare on purpose. If mandate or task authority is not in the graph, aefi says so instead of inventing it.

POST /v1/providers/search
{
  "summary": "12 providers matched “CCTP settlement on Arc”.",
  "confidence": "medium",
  "confidence_reasons": [
    "erc_8183_job_lifecycle",
    "payment_settled"
  ],
  "confidence_model_version": "0.1.0",
  "coverage": {
    "status": "partial",
    "known_gaps": ["authorization_evidence_missing"]
  },
  "evidence": [],
  "result": { "total_matched": 12, "sort_by": "score" }
}

Call aefi the way you call any tool.

Same capabilities on HTTP https://api.aefi.io and MCP server aefi. x402 gating is built in and off by default until you turn it on.

Tool HTTP What you get
search_providers POST /v1/providers/search Semantic + structured provider recall
explain_transaction GET /v1/transactions/{hash} Evidence-backed settlement narrative
verify_payment POST /v1/payments/verify Settlement lookup by tx or payment id
lookup_job GET /v1/jobs/{job_id} ERC-8183 job, parties, outcomes

MCP reference · HTTP reference

Built on the rails Arc is shipping.

System USDC transfers, transaction memos, ERC-8004 identity, and ERC-8183 jobs. When Arc mainnet opens, the graph model is already keyed by chain. Today the live studio reads Arc testnet evidence.

  • System USDC as settlement truth
  • Memos as job / invoice / payout context
  • 8004 identity onto job providers
  • 8183 lifecycle and outcomes
  • Not a chain explorer
  • Not every unbound USDC transfer in the graph
  • Not task-level authority until adapters exist
  • Not a replacement for 8004 / 8183 themselves

Search a provider. Explain a settlement.