This document describes how LogionOS processes customer data when you use the compliance API, proxy, SDKs, and related integrations. It complements our Privacy Policy with implementation-level detail. For contractual terms, see our Terms of Service and, where applicable, your Data Processing Agreement (DPA).
When your application sends a request through LogionOS (for example via the REST API, an SDK, middleware, or the hosted proxy), the full query text is used in memory to run the compliance engine and produce a result for your app. Persistent records on LogionOS infrastructure use a one-way fingerprint of the query—not the raw query text—alongside compliance outcomes and audit metadata, as described in Section 3.
Optional paths (such as BYOK AI Judge or opt-in telemetry) only apply when you explicitly enable them.
Your AI App → LogionOS Proxy / SDK → Compliance Engine → Result
│
▼
Audit Log (hash only)
│
▼ (opt-in)
Anonymized Telemetry
For a typical compliance check, we receive and process:
Our audit and reporting layer is designed to avoid retaining reversible query content. Stored artifacts include:
query_hash — SHA-256 hash of the query; we persist only the first 16 hexadecimal characters as a stable, non-reversible fingerprint for deduplication and audit correlation.We do not store the full query text in this audit store. Raw query content exists only for the duration needed to compute the result unless a separate optional feature (see Section 4) sends it to your own provider.
LogionOS does not route your query to a large language model for the core compliance engine unless you enable BYOK (“bring your own key”) AI Judge or an equivalent customer-configured integration.
When BYOK AI Judge is active, the full query may be sent only to the LLM endpoint you configure (your own API keys and vendor relationship). You remain the data controller for that transmission; LogionOS acts as directed by your configuration. If BYOK is off, query text is not sent to third-party LLM providers by LogionOS for that purpose.
We may offer anonymized telemetry to improve models and rule quality—for example aggregate signal about false positives or coverage gaps. Any use of request-derived signal for model training or product analytics beyond essential service operation is opt-in only, with clear consent in the product or contract. You can withdraw consent according to the controls we provide for your workspace.
Data enters LogionOS through different integration surfaces; the storage and LLM rules in Sections 2–5 apply consistently once a request reaches our compliance engine.
| Mode | What typically flows in | Notes |
|---|---|---|
| REST API | Query text, jurisdiction, headers (auth), JSON metadata | Direct HTTPS; same audit hash and result storage model. |
| SDK (Python / JS·TS) | Same payload fields as the API; wrapped by library | Credentials stay on your side except API keys you configure for LogionOS. |
| Middleware | Intercepted request body or derived prompt text from your stack | Your server forwards to LogionOS; you control logging upstream. |
| Proxy | Traffic mirrored or forwarded from your AI app through LogionOS edge | Path depends on deployment; TLS terminates per your architecture. |
For questions about this document or your deployment, email chris@logionos.com.