Sirveil Exposure Verification API

Sirveil turns "is this person's data exposed on that site?" into one API call that returns an answer you can defend. Independent before-and-after proof if you sell privacy or removal; a risk signal you can price with if you run fraud, KYC or underwriting; a capability that would otherwise cost you a crawler team and a broker registry if you are building a product or an agent. Every response carries the evidence, a calibrated probability, and the version stamps of the logic that produced it - so the answer holds up when a customer disputes it, an auditor asks for it, or a regulator wants the method six months from now. Verification is $0.10 per call and a full scan is $0.35. No minimum, no subscription, no seat count.

POST /api/v1/verify takes an identity and a domain and returns a verdict in one to two seconds (p95 measured at 1,463 ms). The verdict has three states, not two: indexed, not_indexed, and indeterminate. "We could not tell" is a first-class answer and is never disguised as a clean negative. Every response carries the evidence inline - source URL, page title, snippet and retrieval timestamp - with a calibrated probability, the linkage strength behind it, and an explains list stating what was queried, what matched, and what did not.

POST /api/v1/scan runs an exposure sweep for one identity: 12 pinned people-search domains queried by name, plus open-web results, breach and stealer-log records, and username matches. It returns ranked, tiered cards - each naming the exposing site, the fields it exposes, why that matters, and its provenance - plus, where the pieces chain together, a synthesised attack-path card with a severity rating.

A scan also returns a coverage block: how many domains were expected, queried, empty, failed and never reached - naming the ones it did not get to. You are told the shape of the hole, not just what was found.

A synchronous scan takes about two and a half minutes (median 154 s, up to 203 s), longer than most gateways will hold a request. Synchronous is the default; send Prefer: respond-async and you get a job id back instead - submit, poll GET /api/v1/jobs/{id} free of charge, collect the shaped result.

Reproducible and versioned. Pipeline, linkage-weight and calibration versions ride on every response, including every error, so any historical answer can be replayed against the exact logic that produced it - the audit artifact regulated buyers ask for.

The limits of the method are published in the response. Every verdict carries a contract object stating the three states and the boundary of what the answer is worth. A negative means "absent from the search index", not "proven absent from the site" - in the payload, not a blog post.

Coverage is your list, not ours. Verification is domain-addressed: you name the site, with no waiting on a roadmap for a domain to be added. Broker classification and the coverage block are backed by a 548-domain registry derived from the California CPPA public data-broker register plus a curated priority set.

Spend ceilings, not surprise invoices. Every account carries a hard USD ceiling and a per-minute rate limit, both enforced server-side, and ceilings are raised on request. GET /api/v1/whoami returns your plan, remaining quota, ceiling and billing rail - and costs nothing, so you can prove a key works before you spend a cent.

Built for agents as well as engineers. A read-only MCP server ships alongside REST, exposing verification as a single tool with no write path. MCP calls perform a verification's work and are billed at the verification price. OpenAPI 3.1 and Postman artifacts are generated from the live spec, so the docs cannot drift from the code.

All API traffic is encrypted in transit and authenticated with a bearer API key scoped to your account. A synchronous verification or scan does not persist the identity submitted to it - results are computed fresh per call. An asynchronous scan job must hold its input and its answer between your submit and your poll: that job is retained for 24 hours and expires after that.

Scan responses are built field by field from an explicit whitelist rather than filtered, so internal scoring inputs, provider names and cost data cannot leak into a payload by being added upstream. Response identifiers are not derived from the subject.

Subscribe through AWS Marketplace and choose Set up your account. AWS returns you to the Sirveil registration page; enter the email that should own the account and your API key is issued there. Call GET /api/v1/whoami to confirm the key, plan and ceiling - it is free - then POST /api/v1/verify with an identity and a domain for your first verdict.

Documentation, the OpenAPI 3.1 specification and a generated Postman collection: https://ai.sirveil.ai/docs/api