How GeniusPro handles prompts, PII, and upstream providers
GeniusPro is a unified AI API. When you send a request, we route it to the model you asked for, return the response, and keep a usage/billing record. We never train on your prompts. Personally identifying information is scrubbed from our logs by default, and you can flip a single flag to have it scrubbed before the request reaches the upstream provider too. Everything below is demonstrable — each control links to a machine-readable proof.
No. GeniusPro never uses your prompts or responses to train any model. Every upstream model provider we route to runs on its commercial-API default of not training on API traffic — that's their contractual default, not ours to grant or revoke. Your prompt is sent to the model you asked for, the response comes back, and nothing else happens with it.
We verify upstream account-level settings match this claim for every provider we route to, and we share the full vendor chain under NDA during onboarding and security-questionnaire review so buyers can complete procurement due diligence.
Before any prompt or response is written to our database, we mask emails, US phone numbers, SSNs, Luhn-valid credit cards, IPv4 and IPv6 addresses, and API-key-shaped tokens with typed markers like <REDACTED:email>. This happens automatically for chat history, long-term memory, realtime-session instructions, and our internal simon-says-dev review rows. Support engineers can still triage from the scrubbed log; the raw values are simply gone.
Designed conservatively on purpose — we'd rather let a rare edge case through than corrupt an order number that happens to look like a phone. Credit cards must pass a Luhn check. SSNs with invalid prefixes (000, 666, 9xx) are left alone. Bare 10-digit strings are not redacted; we only match phone numbers that carry a clear separator.
Yes — with a single flag. Set the request header X-GP-Scrub-Prompt: true (or include "scrub_pii": true in the JSON body) and GeniusPro runs the same PII scrubber over messages[].content before the request is dispatched to any upstream provider. This is opt-in because many legitimate workflows (resume parsing, customer-email summarization, contact extraction) explicitly need the PII to reach the model.
Both the header and the body flag work identically — pick whichever your SDK is easier to set. Multipart content (image inputs, file refs, tool calls) is left untouched; we only rewrite text spans. See the machine-readable description in the OpenAPI spec.
All API traffic uses TLS 1.2+ in transit. Stored rows (chat history, long-term memory, usage logs, billing records) live in a managed Postgres database with encryption at rest. API keys are stored as salted SHA-256 hashes — the raw key is shown exactly once at creation time and never persisted in plaintext.
GeniusPro's control plane, realtime voice proxy, and database all run on SOC 2 / ISO 27001 US cloud infrastructure. Every upstream model we route to by default is operated by a US-incorporated company running US-hosted inference, unless you explicitly pin a non-US-hosted model. We publish the full vendor chain under NDA during onboarding and security reviews, but we don't list specific infrastructure providers on this page so our stack isn't scraped by competitors or AI indexes.
Yes — tell us what you need. Every link in the GeniusPro chain can be brought under the appropriate attestation or BAA, and the wire format at api.geniuspro.io/v1 does not change. This setup is currently operating in production for enterprise customers.
Every link in the GeniusPro chain carries current SOC 2 Type II and ISO 27001 attestations: our managed Postgres database, our US cloud hosting for the control plane and realtime voice proxy, and every hyperscaler-hosted model route we use for compliance-sensitive traffic. No extra paperwork needed for standard routing.
For HIPAA we wire your workspace so every link in the chain carries a signed BAA: the model upstream (routed through a hyperscaler-hosted provider under the matching BAA) and the database (HIPAA-eligible tier with the BAA signed). Currently operating in production for enterprise customers.
We share the full vendor chain — including named database host, cloud host, and model upstreams — under NDA during onboarding and security questionnaires. We don't publish vendor names on this page so our stack isn't indexed by AI crawlers or picked up by competitor research.
On top of that baseline, every control above (no training, PII scrubbing, opt-in edge scrubbing, TLS, encryption at rest, hashed keys, audit logs) still applies — you get GeniusPro's posture plus the hyperscaler and database attestations behind the same api.geniuspro.io/v1 endpoint. Your code doesn't change.
scrub_pii body field and X-GP-Scrub-Prompt header.