Privacy Policy
How Unii safeguards tenant assets, knowledge files, model provider secrets, and operational decision flows across every stage of intelligence execution.
Summary of Core Privacy Guarantees
Unii is designed for multi-tenant isolation. Your documents, extracted embeddings, and provider credentials belong exclusively to your organization. We never use tenant data to train models, we exclude message bodies from server diagnostic logs by default, and every run enforces strict organization and project scopes.
Platform Scope & Purpose
Unii provides a verifiable intelligence workbench for configuring knowledge corpora, deterministic decision flows, and conversational systems. This Privacy Policy governs all data ingested, processed, or generated when interacting with Unii, including the web workbench, backend REST APIs, and embedded React SDK components.
We prioritize data integrity and deterministic verification. Unii is built so that every customer remains the exclusive custodian of their business logic, documents, and intellectual property.
Tenant Isolation & Boundaries
Unii enforces strict multi-tenant boundaries at the infrastructure, database, and application levels:
- check_circle Scoped Access: Every document search, message read, run execution, and background worker job is restricted to the authenticated organization, project, and subject.
- check_circle No Cross-Tenant Retrieval: Database queries strictly isolate vector search spaces. A retrieval operation for one tenant cannot query, index, or inspect another tenant's vector embeddings.
- check_circle Zero Cross-Contamination: Cache keys and runtime worker memory are isolated per tenant to prevent data bleeding during parallel intelligence runs.
Knowledge Documents & Embeddings
When you upload knowledge files (such as Markdown, PDF, or text documents) to Unii:
Original files are stored in private bucket storage. Document downloads and inspections use short-lived, authorized signed URLs with strict expiry.
Ingestion workers segment text into contextual chunks and generate vector embeddings stored in protected, tenant-indexed database partitions.
Provider Credentials & API Keys
We treat third-party model credentials and platform access tokens with high-grade security protections:
- vpn_key Server-Only Secret Bindings: Foundation model keys (OpenAI, Anthropic, Google Gemini) are stored in server-only encrypted configurations. Secrets are never exposed in browser bundles, DOM states, or client responses.
- key_off Secrets Never in URLs: No secret, credential, or sensitive bearer token is passed via query strings or URL paths.
- tag Irreversible API Key Hashing: Unii platform keys are hashed using SHA-256 before storage. Plaintext keys are presented once at generation and cannot be retrieved later.
Conversations & Diagnostic Logs
Unii handles conversational queries through deterministic grounding pipelines:
visibility_off Redacted Diagnostic Logging
By default, server diagnostic logs record only high-level run metadata (request ID, tenant ID, token count, duration, status codes). Document text and prompt message bodies are intentionally excluded from diagnostic logs to preserve privacy.
When operating in Strict Verification Mode, the intelligence pipeline verifies grounding evidence before returning output. If supporting evidence is insufficient, the model abstains rather than inventing answers or exposing out-of-scope data.
Third-Party Foundation Models
Unii acts as an intelligence orchestrator that proxies execution requests to customer-configured foundation model providers (such as OpenAI, Anthropic, or Google Gemini):
- • Data transmitted during inference is processed under the respective provider’s enterprise API terms.
- • Under standard enterprise agreements, inputs and outputs sent through API endpoints are not used to train or enhance foundation models.
- • Developers using the React SDK interact with the Unii server via scoped embed sessions without exposing permanent provider keys.
Data Retention & Customer Deletion Rights
You retain full control over your stored assets:
Deleting a document instantly removes the file from storage and purges all parsed chunks and vector embeddings from the index.
Deleting a conversation clears its transcripts, grounded citations, and execution traces from the active database.
System database backups are encrypted at rest and retained per our documented retention schedule before automated cycle expiration.
Sessions, Security & Audit Logs
Platform integrity is reinforced by industry-standard technical defenses:
- • CSRF Protection: Cookie-authenticated mutations enforce Cross-Site Request Forgery defenses.
- • HTTP-Only Cookies: Workbench session tokens are stored in secure, SameSite HTTP-only cookies.
- • Immutable Audit Trails: Membership changes, draft snapshot publications, and credential modifications are logged with timestamps and operator identity (excluding secret values).
- • Rate & Budget Controls: Per-project request limits, token budgets, and provider timeouts guard against runaway execution and denial-of-service attempts.
Contact & Compliance Inquiries
For questions regarding this Privacy Policy, data subject access requests, or to report a security vulnerability: