The platform
One engine, the verification layer for AI, behind every item below. Everything listed here works in production today; what is still being built is not on this page.
Ask once, get six independent answers cross-checked into one, and see the measurement behind it.
One question goes to Claude, GPT, Grok, Gemini, Perplexity, DeepSeek at once and comes back as one synthesized answer with a consensus score and a confidence band.
Who it’s for: Anyone with a free account; the Free plan allows 5 questions a day.
After an answer, the six AIs see each other’s replies and can challenge, defend and revise over up to three rounds; you watch each one move toward or away from the consensus.
Who it’s for: Signed-in users, on their own questions: 2 runs to try on Free, 5 a month on Connect, 50 a month on Master with up to 3 rounds.
Every answer’s receipt separates what the models agreed on from where they split, names which model said what, and shows how complete the panel was.
Who it’s for: Everyone: each receipt is a public page you can share.
Attach a PDF, Word file, text or Markdown document (up to 10 MB) or an image, and the question is answered about it; image questions go to the four models that can see.
Who it’s for: Anyone using Multi-Model Chat.
Your last 50 verifications are kept with their receipts; every receipt and deliberation has a share link, and you can mark an outcome on your own answers to improve calibration.
Who it’s for: Signed-in users.
Paste a claim or a transcript; six models say where each claim was first published, every named link is resolved, and each claim is marked sourced, disputed, unsourced or unresolved.
Who it’s for: Signed-in users: 3 a day on Free, more on paid plans. This is sourcing, not a truth verdict.
A nightly measurement of how the six AIs drift, flip under pressure, refuse and cite, on a fixed public set of questions, scored by code rather than by another model.
Who it’s for: Everyone, no account. The data is public JSON too.
A page per model with response rate, latency and agreement with the verified consensus, rebuilt daily, and a status page with each provider’s seven-day success rate.
Who it’s for: Everyone.
For companies whose AI chatbot or support bot answers customers about refunds, pricing, policies or eligibility.
Paste your policy or product info and an answer your chatbot gave; Verify splits the answer into statements and shows which ones your document supports, which it contradicts, and which conditions the answer dropped.
Who it’s for: Businesses running an AI chatbot or support bot. If you don’t run a chatbot, you don’t need Verify.
A check you ran in the free tool can be published as a redacted page for 90 days, so a colleague can see what the chatbot said and what the document says.
Who it’s for: Anyone using the free tool at /verify.
Your document, for example:
What your chatbot said, from:
Add verification to your own app, website or AI tools, self-serve with a key, or run multi-model answers on the engine by request.
POST the question, the drafted answer and the passages it should follow; get a release recommendation (allow, cite required, revise, human review) and the evidence for every statement, with an OpenAPI spec and idempotent requests.
Who it’s for: Developers. 50 checks a month free with an API key; Builder and Team for more.
The same check as two tools (verify_answer and verify_against_document) that Claude Code, Claude Desktop, Cursor, VS Code or any MCP client can call before an assistant states a fact or acts on it. Hosted; nothing to run.
Who it’s for: Developers and anyone using AI tools that support MCP. Every call counts against the same Verify allowance.
A GitHub Action that re-checks your approved answers when a pull request changes the document they came from, comments with the ones that no longer hold, and writes a signed record of the run. Report-only until you choose otherwise.
Who it’s for: Teams that keep approved answers next to their documents. Local mode is free; hosted runs are on Builder and Team.
Create and revoke keys, set expiries, and see this month’s checks and model cost per key; GET /v1/usage returns the same split by API, MCP and Gate.
Who it’s for: Any signed-in account.
GET /api/drift/latest returns the latest nightly run and per-model scores; GET /api/drift/probes returns the probe set. Both are open, with CORS, no key.
Who it’s for: Researchers and developers.
Multi-model answers, deliberation and the receipt inside your own product, set up with us rather than self-serve.
Who it’s for: Builders. Tell us what you are building and the volume you expect; we reply by email.
Verify checks answers against the documents you supply; its verdicts describe what those documents support, not whether a statement is true. Not legal advice. Pricing →
