Private AI Chat: No-Training, History and Hosting Controls
Understand private AI chat in Krater: no-training eligibility, EU-hosted workspace data, incognito controls and the limits of each protection.
Short answer: Private AI chat starts with the service route, the information you disclose and the controls you actually enable. In Krater, Private means no-training eligibility under the configured provider registry. EU-hosted workspace data and incognito or saved-history controls are separate protections with separate limits. Krater does not claim zero retention or end-to-end encryption. Choose approved data and a suitable model before writing the first prompt.
Define private before choosing a chat tool
Businesses often use private to describe several different needs: keeping a project out of a model's training data, limiting what colleagues can access, avoiding a saved conversation or meeting a hosting requirement. These are not interchangeable. Write down which requirement matters for the task, then identify the control or contractual term that addresses it.
Start with Krater's private AI hub for the product boundaries. The guide to models that don't train on your data separates consumer, API and team policies. If the conversation will browse the web, also use the private AI search checklist. A search-enabled chat involves more than a single model response.
A decision table for private chat
| Requirement | What to inspect | Boundary |
|---|---|---|
| No model training on submitted work | The service route, provider terms and relevant badge | No-training is not zero retention |
| No new saved workspace chat | Incognito and history settings | Not a statement about every backend or provider record |
| EU-hosted workspace data | Krater's hosting boundary and external services | Not an all-provider EU-only processing promise |
| Controlled disclosure | Prompt content, attachments and app permissions | A privacy control cannot undo an unnecessary disclosure |
| Organization approval | Account type, team policy and permitted tasks | A personal subscription is not organizational authorization |
What the Private badge means in Krater
The code's private-provider registry contains openai, anthropic, google, mistralai, cohere and x-ai. A model qualifies through its provider prefix. The badge and filter therefore concern the configured provider route, based on an enterprise contract or published no-training policy. They do not claim that every consumer product from those companies has the same training defaults.
The client and shared server registries contain the same six provider IDs. The model picker uses the registry, and the server checks the effective private-only setting for relevant requests. Eligible Max users or teams can restrict choices with Only use Private models. Check that the setting is enabled for the workspace being used rather than assuming that a different account's preference carries over.
Look at the current picker before submitting sensitive work. Model availability can change, and a brand name is not enough to identify a route. If your organization requires a particular contract, region or subprocessors list, ask those questions directly. The badge is a useful no-training eligibility signal, not a replacement for the complete review.
History controls are a different decision
In Settings, open Data and review Save chat history. Incognito or history controls determine whether new chats are saved in your Krater workspace. They should not be described as erasing every operational record held by every service. Decide what you need to preserve for your work and what your organization permits you to save.
Before changing a history preference, make a plan for the result. If the task produces a useful draft, store the approved final document in the correct business system. Do not rely on an unsaved conversation as the only copy of an important decision. Equally, do not export a complete chat into an unapproved shared folder just to keep the answer.
Use less context, not no context
Useful chat does require enough information to answer the question. The goal is to provide the smallest adequate brief. Replace customer names with roles, summarize a dispute instead of uploading the full email archive and remove access tokens from examples. Keep commercially sensitive figures out unless they are essential and the workflow is approved.
Attachments deserve their own check. A document may contain comments, names or background information that you did not intend to disclose. Review the selected file before attaching it and prefer a carefully prepared excerpt when the task only needs a paragraph. A model's privacy eligibility does not decide whether you have authority to share another person's data.
Connected apps expand the boundary
Krater brings 500+ AI models and 1,400+ app connections into one workspace. That does not make every connection appropriate for every conversation. A connected business application has its own permissions and processing terms. Select the sources needed for the task and review any proposed action before allowing it to publish, send or change business data.
EU-hosted workspace data is a hosting statement about Krater. It is not a guarantee that every external provider or app processes all content exclusively in the EU. Krater does not claim zero retention, end-to-end encryption or that a Private badge makes all confidential material safe for every use. These limits should be visible to a team's users, not hidden in procurement notes.
Choose a workspace for the work you can approve
Use the all-in-one AI overview to compare workflows and Krater pricing to review plans and credit allowances. For managed requirements, discuss the complete use case through enterprise. If you are moving from another private-positioned app, compare Venice alternatives and the Venice privacy guide without assuming equivalent contracts.
FAQ
Is private AI chat the same as local AI?
No. Krater's Private badge concerns provider no-training eligibility. It does not say that inference runs only on your device.
Can I use the Private badge instead of reviewing team policy?
No. Verify the service route, the information you may disclose and any organization-specific requirements. The badge answers one part of that review.
Does incognito provide end-to-end encryption?
No. Krater does not claim end-to-end encryption. Incognito and history controls concern saved workspace conversations, not that encryption guarantee.
Policy context, checked November 2026
OpenAI enterprise privacy distinguishes business and API commitments from consumer use. Anthropic's commercial training guidance likewise describes commercial defaults and feedback exceptions. Both sources were checked November 2026. These vendor examples illustrate why the route matters; Krater's own controls are described above without importing unrelated vendor guarantees.