Sign up

COMPARE

Private AI, side by side.
From each provider’s own pages.

What Undergrove, ChatGPT, Claude, Mistral Vibe and Proton’s Lumo say about who can see your conversations, how long they’re kept, training, protection while a model reads them, where the models run and whether they’re open-weight. Every statement about another service links to that service’s own page.

Checked on
Plans
Personal plans, free and paid. Business plans often differ.
Not stated
Where a provider’s pages don’t say, we say so rather than guess.
Beside Undergrove
Undergrove, ChatGPT, Claude, Mistral Vibe and Lumo compared
UndergroveThe workspaceChatGPTOpenAIClaudeAnthropicMistral VibeVibe, formerly Le Chat · Mistral AI26LumoProton
What the provider can seeIn the workspace, not what you write: your browser encrypts each message to a hardware enclave, and our relay passes on ciphertext. We see your account, your IP address (through our host) and each request’s size, time and estimated cost. Our developer API reads requests in memory to check them, and doesn’t keep them.12OpenAI says a limited number of authorised staff and trusted service providers may access conversations: to investigate abuse or security incidents, give support, handle legal matters, or improve models unless you opt out.3Anthropic says chats used to improve Claude are de-linked from your account before review and seen by a small number of staff involved in training. Chats flagged by its safety classifiers may be used for trust and safety.45Mistral’s help centre says authorised team members may access conversations you report, anyone with a shared link can read that conversation, and feedback sends the related input and output.6Proton says it can’t read your stored chat history, which has zero-access encryption. Prompts are encrypted to Lumo’s model servers, which decrypt them to answer.78
How long conversations are keptConversations stay in your browser. Synced and shared copies are encrypted there first, so we store only ciphertext. Tinfoil states nothing persists in its enclaves between restarts.1Until you delete them. A deleted chat is scheduled for permanent deletion within 30 days, unless it’s kept for security or legal reasons. Temporary Chats are deleted after 30 days.91011Deleted chats leave your history at once and Anthropic’s back end within 30 days. With model improvement on, chats may be kept de-identified for up to 5 years; flagged content for up to 2 years.12Until you delete them or your account, then removed after a short grace period; Mistral also cites monitoring and legal retention. Zero data retention isn’t available for chat.13History is kept on your device and synced with zero-access encryption. Proton says it keeps no logs of prompts or replies, and erases them once an answer is made.7
Used to train modelsNo. We keep no conversation content we can read, and the models’ makers never receive your messages.1Yes by default on personal plans. You can turn off “Improve the model for everyone” in Settings › Data controls; Temporary Chats aren’t used.1411If you allow it, under Model Improvement in Privacy Settings, and when a chat is flagged for safety review. Incognito chats aren’t used.155Yes by default on personal plans; you can opt out in settings. Enterprise customers are opted out by default.1617No, Proton says.7
Protected while processedYes, in the workspace: messages are decrypted only inside a hardware enclave whose attestation your browser checks before sending. Your browser verifies Tinfoil’s router enclave; the model’s own enclave is verified by the router, and by your browser where its hardware allows.2Not stated for personal plans. OpenAI’s business pages describe encryption at rest (AES-256) and in transit (TLS 1.2 or higher).18Not stated for the Claude apps. Anthropic has published research on confidential inference with trusted virtual machines, which it describes as early work.19Not stated. Mistral describes encrypted backups replicated across EU zones.20Not stated as hardware-protected. Prompts are encrypted in transit to the model servers, which decrypt them to process them; no enclave is described.8
Where the models runIn Tinfoil’s hardware enclaves, on its confidential GPUs. Tinfoil doesn’t publish the region, and our register says so.12OpenAI’s EU privacy policy says personal data is processed on servers outside the EEA, Switzerland and the UK, including in the United States.21Anthropic’s privacy policy says personal data is transferred to its servers in the United States, or other countries outside the EEA and UK.5In the European Union by default; some features may send data outside the EU temporarily.22On servers Proton controls. Proton says Lumo is built and based in Europe; these pages don’t name a country.7
Open-weight modelsYes. Every model offered is open-weight, and one small model can run in your browser.23Not stated as open-weight. OpenAI’s open-weight models are its separate gpt-oss family, which it calls its first since GPT-2.24No. Anthropic describes keeping its model weights encrypted and secured.19Its default model, Mistral Medium 3.5, has open weights, under a modified MIT licence.25Yes. Proton says Lumo is powered by open-source models, and names them.7

How this was written

Each statement about another service paraphrases that service’s own help pages, policies or announcements, read on 29 October 2026, and links to them. We didn’t use reviews or third-party write-ups. Policies change, and settings can change the answer, so the linked page is the authority. Mistral renamed Le Chat to Vibe in 202626; the same service is described here.

About our own column

It describes what Undergrove’s code does today, and links to the pages that explain it, including the architecture with excerpts of the code. It hasn’t been independently assessed yet. Our host, Vercel, receives ordinary request logs, as the privacy page says.

Sources

  1. Undergrove: Where your data goesOn this site
  2. Undergrove: ArchitectureOn this site
  3. OpenAI Help Center: How OpenAI handles data in consumer serviceshelp.openai.com · checked 29 October 2026
  4. Claude Help Center: Who can view my conversations?support.claude.com · checked 29 October 2026
  5. Anthropic: Privacy Policywww.anthropic.com · checked 29 October 2026
  6. Mistral Help Center: Can other people view my conversations?help.mistral.ai · checked 29 October 2026
  7. Proton Support: Lumo privacyproton.me · checked 29 October 2026
  8. Proton blog: Lumo security modelproton.me · checked 29 October 2026
  9. OpenAI Help Center: Deleting and archiving chats in ChatGPThelp.openai.com · checked 29 October 2026
  10. OpenAI Help Center: Chat and file retention in ChatGPThelp.openai.com · checked 29 October 2026
  11. OpenAI Help Center: Data Controls FAQhelp.openai.com · checked 29 October 2026
  12. Anthropic Privacy Center: How long do you store my data?privacy.claude.com · checked 29 October 2026
  13. Mistral Help Center: Can I activate Zero Data Retention (ZDR)?help.mistral.ai · checked 29 October 2026
  14. OpenAI Help Center: How your data is used to improve model performancehelp.openai.com · checked 29 October 2026
  15. Anthropic Privacy Center: Is my data used for model training?privacy.claude.com · checked 29 October 2026
  16. Mistral Help Center: Do you use my user data to train your Artificial Intelligence models?help.mistral.ai · checked 29 October 2026
  17. Mistral Help Center: Can I opt out of my input or output data being used for training?help.mistral.ai · checked 29 October 2026
  18. OpenAI: Business data privacy, security, and complianceopenai.com · checked 29 October 2026
  19. Anthropic: Confidential Inference via trusted VMswww.anthropic.com · checked 29 October 2026
  20. Mistral Help Center: How do you ensure the security of user data?help.mistral.ai · checked 29 October 2026
  21. OpenAI: EU Privacy Policyopenai.com · checked 29 October 2026
  22. Mistral Help Center: Where do you store my data or my Organization’s data?help.mistral.ai · checked 29 October 2026
  23. Undergrove: ModelsOn this site
  24. OpenAI: Introducing gpt-ossopenai.com · checked 29 October 2026
  25. Mistral AI: Remote agents in Vibe. Powered by Mistral Medium 3.5.mistral.ai · checked 29 October 2026
  26. Mistral Help Center: Le Chat is now Vibehelp.mistral.ai · checked 29 October 2026