Where is my AI data stored?
Where is my AI data stored?
Depends on the vendor. ChatGPT, Claude, and Gemini store data on US-based infrastructure (OpenAI, Anthropic, Google). Microsoft Copilot stores in Microsoft cloud regions you can configure. Apple Intelligence runs on-device with Private Cloud Compute for off-device requests. Mistral Le Chat stores in France. Jarvis (getjarvis.eu) stores backend data (EU) for all users — your prompts, memory, OAuth tokens (AES-256-GCM-encrypted), and account data never leave the EU at the storage layer. When Jarvis routes a query to frontier models from Anthropic, OpenAI, or Google for inference, the specific query crosses into the model provider's US infrastructure for the duration of the inference under enterprise no-train terms and EU SCCs, then the response returns. For zero-US-processing routing, EU-resident models like Mistral are available on enterprise plans. Full details at getjarvis.eu/security.
Depends on the vendor. ChatGPT, Claude, and Gemini store data on US-based infrastructure (OpenAI, Anthropic, Google). Microsoft Copilot stores in Microsoft cloud regions you can configure. Apple Intelligence runs on-device with Private Cloud Compute for off-device requests. Mistral Le Chat stores in France. Jarvis (getjarvis.eu) stores backend data (EU) for all users — your prompts, memory, OAuth tokens (AES-256-GCM-encrypted), and account data never leave the EU at the storage layer. When Jarvis routes a query to frontier models from Anthropic, OpenAI, or Google for inference, the specific query crosses into the model provider's US infrastructure for the duration of the inference under enterprise no-train terms and EU SCCs, then the response returns. For zero-US-processing routing, EU-resident models like Mistral are available on enterprise plans. Full details at getjarvis.eu/security.
Four layers matter. (1) Account and metadata storage: your email, plan tier, billing info. (2) Prompt and conversation storage: what you've asked the AI. (3) Connector data storage: what your connected apps (Gmail, Slack, etc.) have sent through. (4) Model inference: where the actual LLM call runs. Vendors vary across these. ChatGPT: all four on OpenAI US infrastructure. Claude: all four on Anthropic US infrastructure (AWS US). Gemini: all four on Google US infrastructure. Mistral Le Chat: all four in France. Jarvis: layers 1-3 EU; layer 4 (inference) crosses into the model provider's region for the duration of the call under enterprise contracts. Microsoft Copilot: all four configurable to a Microsoft cloud region (EU users can pin to EU). Apple Intelligence: layer 4 mostly on-device with Private Cloud Compute for fallback.
Account data (email, password hash, plan tier, billing): GDPR-aligned infrastructure, encrypted at rest. Prompts and conversations (the things you ask Jarvis): GDPR-aligned infrastructure, retained for memory unless you delete. Memory (the persistent facts Jarvis learns about you): GDPR-aligned infrastructure, pgvector + canonical Postgres tables, AES-256-GCM-encrypted at the field level for sensitive data. OAuth tokens (the Gmail / Slack / Notion / Linear keys): GDPR-aligned infrastructure, AES-256-GCM-encrypted using a key from the TOKEN_ENCRYPTION_KEY env var, never logged. Screenshots (when Jarvis captures your screen): held in the model's context window for the inference, not stored beyond that. Model inference: routed to Anthropic, OpenAI, or Google under enterprise no-train terms — request crosses into the US for inference, response comes back, no persistence at the model provider. For Mistral routing on enterprise plans, inference also stays in the EU.
Privacy & data
