Frequently asked questions

Every gray-zone sample you submit is scoped to your tenant — it's only ever used to train and evaluate your own verifier, never shared across tenants. Passwords are bcrypt-hashed, API keys are stored as a one-way hash, and all traffic runs over HTTPS. See the full Privacy Policy for details.

Yes — CacheVerifier doesn't replace your cache or do similarity search itself. You call /v1/verify on candidates your own cache lookup (GPTCache, LangChain, a self-built Redis lookup) already produced, specifically the gray-zone ones where similarity alone isn't a confident enough signal.

The similarity range where a cache hit might be right or might be wrong — high enough that a plain threshold would call it a match, not high enough to be confident. That's the range where a verification call actually changes the outcome; outside it, a plain similarity score is usually fine on its own.

Any of them — /v1/verify is a plain REST call that takes a (query, candidate_answer) pair, so it's backend-agnostic by design. integrations/gptcache_adapter.py implements GPTCache's own SimilarityEvaluation interface as a ready-made drop-in; LangChain or a self-built Redis cache just call the REST API directly.

The API enforces a hard floor of 20 rows. Our own testing across four independent datasets found fine-tuning can be a net negative below roughly 500 rows on some data — see the Research page for the actual numbers. That's exactly what the Health Check Report and the built-in guidance on that page are for: telling you whether your current sample is ready before it becomes your tenant's live verifier.

Yes, with too little data or too-noisy labels — this is a real, measured failure mode, not a hypothetical (see Research). It's also why the Health Check Report flags small samples and near-ceiling baselines instead of just reporting a number, and why drift monitoring exists: to catch a verifier that's drifted out of step with current traffic.

Today CacheVerifier is offered as a hosted service only — there isn't a packaged self-host distribution. The underlying verifier/fine-tuning code is architected to be backend-agnostic, but running your own instance isn't a supported product path right now.

None yet — the service is in beta and provided without a formal SLA. We'll publish one when we're out of beta; see the Terms of Service for the current status.

Everything is free during beta, so there's nothing to lose access to by not paying. When paid plans launch, self-serve verify and fine-tuning stay free — see Pricing for what's planned to be paid versus free.