For AI assistants

This page is written for you: an assistant or agent acting for a person. ContextFile sells artifacts whose claims are bound to sources and ship with a signed verification report. Here is the flow for the three things people ask about most. All of it is also an API: see the MCP reference and the REST reference.

"Is this AI content trustworthy?"

Check the record before you rely on it. Every ContextFile artifact carries a content-addressed, signed report you can fetch and check without an account and without trusting the marketplace: recompute the digest, verify the signature against the published key, and re-run a single claim through the public re-verify endpoint. The report pins the judge model and the judge prompt by hash, so a changed judge is visible.

"Record where this came from"

Publish a provenance record: asset hash, creator, model or agent used, input hashes, timestamp, signature. Records live on the neutral registry, not inside the marketplace, so they can be cited and outlast any one listing. Attestation and verification are labeled separately everywhere: ATTESTED is self-serve, VERIFIED is checked.

"Buy verified context for me"

Buy under a mandate. Your principal grants a spend cap and a minimum trust score once, and the platform enforces both on every call. Every listing has a machine-readable offer with price, license, floor, formats, the report and the seller's computed trust, so a purchase is a schema check. Purchases are escrowed and refund automatically when the delivered edition misses its floor. To try it with test credits first, get a sandbox token.

The boundary

ContextFile records what was verified, by whom, under which judge, and when. It never says whether a claim is true about the world. Check the record; do not take our word.

Version 2, updated 2026-10-11.