Short answer: "No training" and "no retention" are two different promises. The business plans from OpenAI, Anthropic, Microsoft and Google all say they do not train on your business data by default, but they keep your chats as part of the product. Fewer tools say they keep nothing at all: local apps on the local-model path, and some privacy-first hosted services. Faradex holds files and conversations in memory only.
What is the difference between no training and no retention?
No training means the provider says it will not use what you type to improve its AI models. No retention means it does not keep what you type once the answer comes back. A tool can promise the first without the second. A chat that is kept, even one never used for training, still exists on systems you do not control.
Tools that say they keep nothing, or close to it
| Tool | What it says it keeps | Training | Team use |
|---|---|---|---|
| Faradex | Files and conversations held in memory (RAM) on your Workspace's dedicated infrastructure, with no backups; gone when you reset. Usage logs record audit details only, never content. | Never used to train any model (faradex.ai/security) | Shared Workspace |
| LM Studio, local model | "Nothing you enter into LM Studio when chatting with LLMs leaves your device." Its cloud models are a separate path. | Local path: nothing leaves your device | Teams plans for its cloud |
| Jan, local model | "runs 100% offline on your computer"; it can also connect to cloud models, whose rules then apply | Local path: nothing leaves your computer | No team product found |
| AnythingLLM Desktop | Runs "entirely on your computer"; the company also sells a cloud version | Local path: nothing leaves your computer | Single user |
| Ollama, cloud plans | "Your prompts are never stored or trained on." It also runs models locally. | "never ... trained on" | Team plan, early access |
| Proton Lumo for Business | Proton says Lumo keeps no record of your chats, and zero-access encryption means only you can read your chat history | None "except with your express permission in giving anonymized feedback" for one model (Proton) | "Administrative tools" |
| Tinfoil Chat | "We do not retain API prompt or response content after the response is returned", apart from temporary caching in encrypted memory; "Chat may store encrypted backups for sync and history" | No, per Tinfoil's FAQ | No team plan seen |
For local apps, the privacy claim holds only when you use a local model. Each of them can also connect to cloud models.
Tools that keep data for a set time
| Tool | What it keeps | Training |
|---|---|---|
| ChatGPT Business | Admins "can control how long your data is retained"; deleted chats removed "within 30 days", with exceptions | "By default, we do not use your business data for training our models." |
| Claude Team | "we retain your chats and coding sessions in the product"; deleted chats removed "within 30 days". Team plans "are not ZDR-eligible" (Anthropic). | "No model training on your content by default" |
| Microsoft 365 Copilot | "The stored data includes the user's prompt and Copilot's response"; admins can "set retention policies" | "aren't used to train foundation LLMs" |
| Gemini in Google Workspace | "Admins can control whether conversations with Gemini are saved and for how long before they are automatically deleted" | "not human reviewed or otherwise used for Generative AI model training outside your domain without permission" |
| Kagi Team | Threads "temporary (threads expire 24 hours after the last message) or permanent"; frontier model providers keep "30 days", some open models "Not stored" | Kagi says never, by Kagi or the model providers (Kagi) |
| Duck.ai | Strips personal metadata before sending; providers delete "at most within 30 days, with limited exceptions" | Providers agree not to use prompts to improve their models; free, for people 18 or older, no business plan found |
ChatGPT Enterprise and Claude Enterprise both say they do not train on your business data unless you opt in.
Every fact above comes from the vendor's own page, checked November 10, 2026.
How to check a tool yourself
- Find the vendor's privacy or data page, not just its marketing page.
- Look for two separate answers: does it train on your data, and how long does it keep it?
- Check what happens to deleted chats, and to "temporary" or "incognito" chats.
- For apps that run locally, check whether a cloud model is switched on.
- For a business, check what an admin can control.
Which one fits your business?
Keeping nothing usually costs something: local apps need a capable computer for each person, and hosted private services cost more than the free plans. See What private AI costs and Local or hosted private AI. Looked at Claude Enterprise and hit the 20-seat minimum? See Claude Enterprise alternatives for small firms.
This guide is general information, not legal advice.