How FyneDesk handles your data

1. Where your data lives

FyneDesk application data is stored in Supabase in the us-west-2 AWS region (US-West / Oregon, USA). We do not offer EU data residency today and do not imply that customer workspaces are stored in the EU.

Supabase publishes that each project is deployed to one primary region, and lists West US (Oregon) / us-west-2 as a specific region. Supabase also states that customer data is encrypted at rest with AES-256 and in transit via TLS.

2. How tenants are isolated

FyneDesk is a multi-tenant application. Customer records are scoped by organization_id, and tenant isolation is enforced in Postgres using Row Level Security (RLS) policies. The database is the boundary; the user interface is not the source of truth for tenant separation.

For server-side actions that run with elevated privileges, FyneDesk resolves the organization from the authenticated user or from the trusted connection context on the server side. We do not rely on a browser-supplied organization id to decide what a user may read or modify.

3. What the AI sees and when

FyneDesk AI is controlled by workspace administrators in Settings and can be switched off anytime. AI acts when a team member invokes an AI action or when an administrator has trained, tested, configured, and enabled autonomous resolution. FyneDesk does not continuously stream workspace content to AI providers.

For each AI action or eligible autonomous-resolution attempt, FyneDesk sends only the context needed for that action. Depending on the workspace settings, that context can include the current ticket, relevant knowledge base text, similar resolved tickets, asset context, or calendar and maintenance context. Administrators control which knowledge sources are enabled and train the system by defining guidance and boundaries, testing expected behavior, and setting confidence and escalation rules. Autonomous resolution answers and closes only requests that meet the administrator's configured requirements; other requests are handed to a person.

FyneDesk uses Anthropic Claude for AI text generation and analysis. It uses Groq only for lightweight sentiment analysis and spam or abuse checks. For embeddings and retrieval, production uses Voyage AI voyage-3.5.

4. Subprocessors

These are the main subprocessors involved in operating FyneDesk. Some receive all app data because they host or deliver the service; others receive data only when a feature is used.

A named subprocessor list is available at fynedesk.ai/subprocessors.

5. Model training

FyneDesk does not use customer workspace content to train FyneDesk-owned models.

Anthropic documents that commercial/API prompts and outputs are not used for model training unless a customer opts in or gives express permission. Anthropic also documents a standard retention period for commercial/API users.

Groq's Services Agreement states that inputs and outputs are not used to train or fine-tune models unless the customer explicitly permits it. Groq also documents that inference data is not retained by default, although inputs and outputs may be logged for up to 30 days for platform reliability or suspected-abuse investigations unless the applicable data-retention control is enabled.

Voyage AI's public terms allow model-training use unless a customer opts out, and its FAQ documents that customers can opt out of storage and future model training for Voyage-hosted model API endpoints. FyneDesk's Voyage account is opted out, so data sent to Voyage is not used for future model training.

6. Retention and deletion

Active workspace data is retained while the organization uses FyneDesk, unless an admin or authorized user deletes specific records earlier. FyneDesk automatically deletes workspace data after six consecutive months of inactivity. When FyneDesk deletes an organization, organization-scoped application records are removed from production tables through explicit deletes and database foreign-key cascades.

AI operational data has narrower retention controls. The current retention sweep removes eligible AI retrieval request logs after 90 days, grounding and confidence logs after at least 90 days, eval runs after 365 days, closed AI security events after 540 days, and orphaned AI chunks after 30 days. Live AI chunks tied to existing source documents are retained while the source document exists so search and retrieval continue to work.

Some billing, security, audit, abuse-prevention, and provider-side records may be kept longer where needed to operate the service, meet legal obligations, resolve disputes, or prevent abuse. For a formal deletion commitment, request a signed DPA.

7. Who to contact and DPA availability

FyneDesk is operated by FyneDesk LLC in Elk Grove, California, United States. For security, privacy, DPA, or deletion questions, contact support@fynedesk.io.

Our current Data Processing Agreement forms part of the Terms of Service. A signed counterpart is available on request.