For those who value their privacy, the growing presence of AI personal assistants can be unsettling. Using these tools typically means handing over personal details, which are then stored by the company behind the model. Given that OpenAI has begun experimenting with ads, it’s not hard to picture the same kind of data harvesting that powers platforms like Facebook and Google making its way into your chatbot exchanges.

A fresh initiative, introduced in December by Signal co-founder Moxie Marlinspike, offers a glimpse of what an AI service built around privacy could be. Confer is crafted to mirror the look and feel of ChatGPT or Claude, but its backend is structured to prevent data collection, applying the same open-source discipline that has earned Signal widespread trust. Your chats with Confer cannot be used to train the model or serve ads, simply because the host never gains access to them.

For Marlinspike, these safeguards are a direct answer to the deeply personal nature of such services. “It’s a form of technology that actively invites confession,” he explains. “Chat interfaces like ChatGPT know more about people than any other technology before. When you combine that with advertising, it’s like someone paying your therapist to convince you to buy something.”

Achieving this level of privacy relies on multiple systems working together. First, Confer encrypts messages sent to and from the system using the WebAuthn passkey standard. It’s worth noting that this approach functions most smoothly on mobile devices or Macs running Sequoia, though it can also be configured on Windows or Linux with the help of a password manager.

On the server side, all of Confer’s inference processing occurs within a Trusted Execution Environment (TEE), supported by remote attestation systems that confirm the environment hasn’t been tampered with. Inside this secure space, a selection of open-weight foundation models handles whatever query comes through.

This architecture is significantly more complex than a standard inference setup—which is already quite involved—but it fulfills Confer’s core commitment to users. As long as these protections remain active, you can engage in sensitive conversations with the model without any information escaping.

Confer’s free tier caps usage at 20 messages per day and five active chats. For users willing to subscribe at $35 a month, unlimited access is provided, along with more advanced models and personalization features. That price is notably higher than ChatGPT’s Plus plan—but then again, privacy rarely comes cheap.