Powerful AI models. More control over what you share.
Meet Basin One: a workspace designed to bring supported AI models together, help you choose the right model for your task, and make privacy part of how you work. Whether you are drafting an email, exploring an idea, or preparing professional work, you deserve useful AI and clearer information about how your requests are handled.
01
Different tasks call for different strengths. Instead of switching between separate AI tools, Basin One is designed to give you access to supported models in a single workspace. Choose a model for the work in front of you, whether that means writing, research, brainstorming, or analysis. Available models should be confirmed against the released product.
02
We believe the way an AI platform handles your information matters as much as the answers it produces. Our privacy approach is designed to reduce unnecessary exposure of recognized sensitive details before eligible requests are processed.
03
A multi-model workspace can involve more than one AI provider. We want users to understand when external providers may process requests and why provider policies, logging, and retention still matter.
Privacy does not begin and end with a settings toggle. It involves the entire journey of a prompt, from the moment you submit it to the moment a model responds. Our proposed initial privacy architecture follows four steps; the specific capabilities below require production verification before they can be described as active.
You enter a question or task in Basin One. Like any AI prompt, it may contain personal information, business context, or details you did not intend to disclose.
The planned privacy gateway uses Presidio-based detection and anonymization to recognize certain personal identifiers, alongside checks for recognizable secrets such as credentials. Depending on the request and policy, information may be masked or processing restricted.
Eligible requests may be routed to approved external AI model providers. Where masking is applied, the provider may receive transformed text instead of the original recognized identifiers. This is still external processing, not private inference.
The selected model processes the information it receives and returns an answer. The information left in the request, provider processing terms, and applicable retention policies still affect privacy. For example, a prompt that includes a client name and email address may be transformed to use placeholders before an approved model receives it. This can reduce exposure of those identifiers, but it does not guarantee that the remaining text is anonymous or free of confidential information.
A useful privacy feature should come with a clear explanation of its limits.
| Planned V1 safeguards | Important limitations |
|---|---|
| Recognition and masking of certain personal identifiersUsing a Presidio-based approach. | Unusual identifiers, indirect clues, or contextual information may be missed. |
| Secrets scanning and gateway checksBefore model or tool requests leave the workspace. | Trade secrets, proprietary strategies, and unpublished ideas may not look like recognizable credentials. |
| Policy-based routingTo approved external AI providers. | Approved providers may still process transformed prompt content under their applicable terms. |
| Encrypted storage and conversation controlsIn the proposed architecture. | Exact retention periods, deletion behavior, exceptions, and availability require confirmation. |
Masking is not the same as complete anonymity. Nor does a commitment not to train on customer data necessarily mean that a provider retains no information. More advanced concepts such as semantic trade-secret detection, reversible token vaults, self-hosted private inference, and confidential computing should not be assumed to be available.
If you are handling regulated, contractual, or client-confidential material, check that AI use is authorized before sharing it. Privacy safeguards support responsible decisions; they do not replace them.
Brainstorm, write, learn, and organize ideas in one place. Whether you are preparing a message or exploring a topic, use AI without sharing more personal information than the task requires.
Move between proposals, content drafts, research, and problem-solving with access to different AI models. When work involves clients, consider which names, contact details, or other identifiers can be removed.
Client documents, business plans, source code, and intellectual property deserve extra care. Assess whether the task is permitted, minimize the material shared, and do not assume that automated masking detects every confidential detail.
One AI model may be useful for drafting, while another may be a better fit for analysis or a complex question. Basin One is designed to make supported models accessible from one workspace so you can choose based on the task rather than the platform you happened to open first.
Model choice also affects data handling. Depending on the selected model and routing configuration, an external provider may process the request. The live product should identify supported providers and explain relevant processing terms. Training restrictions, retention practices, and logging are separate questions worth checking.
Explore a more flexible way to work with AI, with multiple supported models and a deliberate approach to privacy. Join the launch list to hear about access and next steps.
Look at how information is collected, checked, transmitted, processed, stored, and deleted. Detection, masking, controlled routing, and private processing environments provide different levels of protection. No single feature guarantees complete privacy.
The proposed initial architecture allows eligible requests to reach approved external model providers. When masking applies, providers may receive transformed content. The exact providers, endpoints, and processing terms must be confirmed for the released service.
Masking may miss indirect identifying clues, unusual personal details, confidential business context, trade secrets, and intellectual property. Removing information can also reduce the context a model needs for a useful answer.
We do not equate masking with full anonymity. Replacing a name or email address may reduce direct identification, but other details can still identify a person or organization.
Conversation deletion and retention controls are part of the proposed initial requirements. Their actual availability, timing, and exceptions need to be confirmed against the released product and privacy documentation.
Understanding how AI handles information helps you choose tools and settings that fit your needs. Explore our guides for practical answers:
At Basin One, we believe powerful AI should be easier to access without losing sight of what happens to your information. We are bringing multiple models into one workspace and putting thoughtful data handling at the center of our approach. Choose the AI that fits your task, understand the privacy considerations, and move forward with greater confidence.