Overview
What VIGIL is, the problem it solves for agents paying through x402, who it is for, and how the pieces fit together.
VIGIL is quality control for agent payments. It tests every x402 endpoint, scores it onchain, and holds an agent's USDC in escrow until the response checks out. This page explains the problem, who VIGIL is for, how its components fit together, and what it deliberately does not do.
The problem: agents pay blind
AI agents on Base pay for APIs through x402. The flow is simple: the agent requests a resource, the server answers HTTP 402 "Payment Required" with a price, the agent pays USDC, and the server returns the response.
The flow has no quality step. Once the USDC moves, the agent has no recourse if the response is empty, if the JSON does not match the schema the endpoint advertised, if the body is truncated, or if the amount charged is not the amount quoted. The money is gone, and the agent often cannot tell that anything went wrong until a downstream task fails.
Nobody neutral watches the endpoints. Providers publish their own status pages. Agents have no shared memory of which endpoints are reliable. Every agent learns the same lessons alone, one failed payment at a time.
VIGIL adds two things to x402: a neutral, staked network that continuously measures endpoints and publishes the results onchain, and an escrow path that makes a payment conditional on the response passing checks.
Who VIGIL is for
Agent developers
You build agents that call paid APIs. VIGIL gives you a public Endpoint Score (0–100) for every measured x402 endpoint, free to read, so your agent can choose which endpoints to trust. When it pays, it routes the payment through the VIGIL Facilitator instead of paying the endpoint directly. USDC is held in escrow and released only if the response passes the schema, size, deadline and price checks. If it fails, the refund is automatic. See Agent developers.
x402 providers
You run paid endpoints. VIGIL is a drop-in x402 facilitator, so your server does not change. A good Endpoint Score is public proof of reliability that agents can act on, and the escrow path means agents can pay you with less hesitation. An optional subscription adds a "VIGIL Verified" badge. See Providers.
Watchers
You stake $VIGIL and run Probes. Probes call paid x402 endpoints continuously and measure uptime, latency, schema validity and quoted price versus charged price. Each result is published onchain as an Attestation. Watchers earn 50% of protocol fees, vote on disputes, and are slashed for publishing false attestations. See Watchers.
How the pieces fit
Watchers run Probes against paid endpoints and publish Attestations to an append-only registry on Base. A deterministic aggregator turns those attestations into an Endpoint Score per endpoint and writes it to the ERC-8004 Reputation Registry, where any agent or developer reads it for free. Separately, when an agent pays through the VIGIL Facilitator, the Facilitator locks the agent's USDC in the escrow contract, forwards the request, checks the response, and instructs the contract to release the funds to the provider or refund the agent. Each escrow verdict is also recorded as an attestation, so paid traffic and probe traffic feed the same score. Contested results are settled by a stake-weighted vote of Watchers.
| Component | What it does |
|---|---|
| Facilitator | Drop-in x402 facilitator. Locks USDC, forwards the request, runs the four checks, submits the verdict. |
| Escrow contract | Holds USDC on Base. Releases to the provider on pass, refunds the agent on fail or timeout. |
| Watchers and Probes | Staked operators whose probes call endpoints every 60 s from at least 3 regions. |
| Attestation registry | Append-only record on Base of every probe result and every escrow verdict. |
| Score aggregator | Weighted mean of attestations with time decay and stake weighting, producing a 0–100 score. |
| ERC-8004 adapter | Writes the score, attestation count, last block and content hash to the Reputation Registry. |
| Dispute module | Lets staked Watchers contest a result; resolves by stake-weighted vote and slashes false attestations. |
What VIGIL is not
- Not a marketplace. VIGIL does not list, sell or route to endpoints. It measures the endpoints an agent already uses and secures the payments it already makes.
- Not an oracle for arbitrary data. Attestations state whether a response met the declaration in its own 402 quote. They do not vouch for the truth of the content.
- Not a wallet. The agent keeps its own keys and its own USDC. VIGIL only holds funds inside a single escrowed call, for seconds, under contract rules.
- Not launched as a token. $VIGIL is not launched. The token exists so Watchers have skin in the game and slashing means something. Launch details will be announced only on x.com/VigilOnBase. Any other address is a scam.
- Not deployed or audited. The contracts described in these docs are not yet deployed and not yet audited. Every address shown is a placeholder.
Where to go next
- Architecture: the components, the data flow, and what lives onchain versus offchain.
- Escrow flow: the escrow state machine, timeouts, refunds, disputes and fees.
- Checks and scoring: the four checks, reason codes, attestation format and the score formula.
- Agent developers: route an agent's x402 payments through VIGIL.
- Providers: what a good score requires and how the "VIGIL Verified" badge works.
- Watchers: stake, run Probes, earn fees, vote on disputes.
- ERC-8004: how VIGIL uses the Identity and Reputation registries.
- Security and risks: trust assumptions, known limitations, and the current audit status.
- Token: the utility of $VIGIL and why it exists.
- Glossary: every term used in these pages.