CAPAS is a deterministic gate, not a model. There is no language model in the decision and nothing to train, so your claims are never used to train anything. This page states plainly what is true today and what is still on the roadmap.
pip install capas-claim-gate. The gate runs in-process on your machine; evidence never leaves it.
python3 capas_mcp.py as a local tool for your agent. Same in-process gate, no outbound calls.
A managed endpoint for signed certificates. Used only if you choose it; auth-gated and isolated per pilot.
source_urls for provenance verification, CAPAS fetches them to confirm they are recoverable and hash-match. Supply no URLs → no network calls in the gate path.No. CAPAS is deterministic — there is no model in the decision path and no training step. The same payload always yields the same verdict.
Not in library or MCP mode — the gate is in-process. The only exception is opt-in provenance verification (if you include source_urls, those URLs are fetched). The hosted API is used only if you send a payload to it.
Certificate issuance is auth-gated (Authorization: Bearer / X-API-Key via CAPAS_API_KEY). Certificates are signed (HMAC-SHA256, content-addressed) so any tampering is detectable on verification. The certificate store path is configurable (CAPAS_DATA_DIR); retention is defined per pilot.
Yes. Run the library or MCP server inside your own environment on redacted or full records, with no source URLs, and nothing is transmitted. Pilots can begin on synthetic samples or customer-provided redacted payloads.
The local gate writes no telemetry. The hosted API logs are operational only and configurable per deployment; no payload content is required to be retained.