UVOX questions, answered.
General product, pricing, security and integration information without disclosure of proprietary engine methods.
What does UVOX do?
UVOX reduces eligible OpenAI and Anthropic API costs through safe context optimisation, provider-native caching, automatic fallback and per-request economics reporting.
Which engine modes are available?
Auto is the recommended combined beta route; Maximum Savings uses the tenant-configured beta confidence and reduction limits; Zero-Loss Cache preserves complete request content; Disabled sends the request directly without UVOX optimisation.
Does UVOX change my prompt?
Zero-Loss Cache and Disabled modes do not remove prompt content. Auto and Maximum Savings may omit historical messages that the router determines are redundant or unrelated. System and developer instructions and recent messages are preserved. Selected relevant tool definitions are forwarded unchanged; unsafe tool-history workflows use complete-context fallback.
Can the model output change?
Yes. Any context reduction can change the model output. Customers requiring complete prompt identity should use Zero-Loss Cache. Combined modes automatically fall back when configured safety or confidence requirements are not met.
What happens when the combined route is unsafe or rejected?
UVOX sends the complete original request through Zero-Loss Cache. If provider cache metadata is also rejected, UVOX sends the complete original provider request without cache metadata.
Does UVOX work with provider caching already enabled?
Yes. UVOX coordinates supported OpenAI and Anthropic provider-native caching. Existing provider caching does not need to be disabled.
How are savings measured?
UVOX reports context tokens avoided, provider input, cache-write input, cache-read input, actual provider cost, estimated baseline and combined cost reduction separately. These percentages must not be added together.
Are savings guaranteed?
No. Savings depend on provider, model, prompt length, context relevance, repeated prefixes, cache eligibility and provider pricing. Test your own traffic before production use.
Does UVOX store prompt bodies in normal logs?
No. Normal production logs store operational and usage metadata, route and fallback evidence, not prompt or generated-answer bodies. Requests are still processed transiently and forwarded to the selected provider.
Have another question?
Contact UVOX Support without requesting confidential engine information.
Contact supportZero-risk verification questions
What exactly does zero-risk mean?
In Zero-Loss Cache Mode, UVOX does not rewrite/delete prompt content, reuse a semantic cached answer, switch provider/model or substitute a stored response. Provider-supported cache metadata may be added.
Can I test the claim myself?
Yes. The browser Proof Lab and downloadable verifier use your own keys, model and prompt. They run direct and UVOX requests, compare hashes and provider usage, and create an evidence package.
Are generated responses expected to be identical?
No. A nondeterministic model can produce different wording for identical prompts. The proof checks prompt content, provider, model, generation route, response IDs and usage—not exact response text.
Does UVOX store my Proof Lab keys or prompt?
No API key or full prompt content is written to the proof database or evidence package. Only hashes, usage, costs, route status and optional anonymised metrics are stored.
What does the 64.7% figure mean?
It was the median cache-only warm-cost reduction in a four-case Anthropic staging benchmark. It is not a guaranteed customer result and does not apply to every workload.