1. Inputs
A response may use the current prompt, selected history, system instructions, permitted files, knowledge materials, and workspace settings. Unrelated data should not be included without need.
2. Processing modes
- Local — processing within a defined local environment;
- Hybrid — separate stages run in different permitted environments;
- External — content is sent to a specifically permitted external provider.
3. Provider policy
An external route must be enabled by an administrator and allowed by server policy. Model-selection UI cannot bypass policy. Country, data categories, and transfer status are published after registry verification.
4. Files, RAG, and knowledge
Files are indexed and retrieved only within an authorized workspace. Embeddings may preserve characteristics of source content and are treated as customer data.
5. Logging and observability
Secrets, full prompts, and file content must not enter logs by default. Diagnostics use minimal technical metadata and correlation identifiers.
6. Model training
The public site makes no universal no-training/training claim for external providers without confirmed contract and registry evidence. Terms depend on provider and deployment.
7. Output review
Model output may be wrong. Verify sources, calculations, code, and conclusions, particularly for high-impact use.
8. Retention and deletion
History, files, indexes, caches, and backups follow an approved deletion chain and schedule. Specific periods are not claimed until the technical matrix is verified.
Document control
- Version
- 1.0
- Effective date
- 2026-08-23
- Internally approved by
- Уполномоченный представитель ООО «Виксора»
- Approval date
- 2026-08-23
- Next review
- 2027-08-23
- Contact
- privacy@yasnora.ai