Legal
AI & Data Processing Policy
How ORQVION Online Beta processes online, offline, and BYOK AI requests — without permanent content storage on ORQVION servers.
Last updated: 15 September 2026
This AI & Data Processing Policy supplements the Privacy Policy for ORQVION Online Beta.
It is product-accurate documentation for customers and operators. It is not legal advice. A signed DPA or MSA controls if it conflicts with this page.
1. Product mode
| Mode | Online Beta status | Where inference runs |
|---|---|---|
| Online (AUTO default) | Default Ready path | Desktop → provider APIs with entitlements from license server (BYOK keys local) |
| BYOK | Optional | Desktop → your provider (your key in keychain) |
| Offline (ORQVION AI + GGUF) | Shipped | Your machine via llama.cpp — not the Ollama app |
OpenAI is not a required core provider. Offline stack is optional at first launch but ships in the same desktop binary.
2. What ORQVION stores
| Data | Stored by ORQVION? |
|---|---|
| Prompts / completions content | No permanent store |
| Source code / project files | No |
| Online entitlement routing metadata | License server — plans, quotas, feature flags |
| Operator-configured provider credentials (if any) | License server secrets vault — not your BYOK keys |
| BYOK keys | Local OS keychain / encrypted store — not uploaded as product data |
| Usage metering | Token counts, hashes, status, latency — not bodies |
| License / account | Entitlement and identity metadata |
The license server is control plane only. It does not receive your repository, index payloads, or chat transcripts for permanent storage.
3. Upstream providers and inference paths
When AUTO or ONLINE routing sends a request to a cloud model:
- Inference runs between your desktop and the provider API (entitlement path and/or BYOK).
- Third-party providers process data under their terms and retention rules.
- ORQVION may configure allowed models and health routing on the license server; that configuration does not create a prompt archive on ORQVION infrastructure.
When OFFLINE routing uses ORQVION AI + GGUF, inference stays on your machine via llama.cpp (not the Ollama application).
4. Customer controls
- Privacy / telemetry consent in the Ready wizard and Settings.
- Disable BYOK providers.
- Enterprise policy can gate models, tools, and sandboxes.
- Delete local chats/indexes by clearing app data.