Wisp, a new project built out of the Lido ecosystem, is trying to make its privacy guarantee mathematically verifiable rather than just contractually stated.

The waitlist for Wisp opened at usewisp.io, giving users early access to what the team describes as a confidential AI assistant that cannot read or retain user data by design, not just by policy.

What Wisp actually does, and why the architecture matters

Wisp runs its large language models inside Trusted Execution Environments, or TEEs. Your prompt goes in encrypted, gets processed inside the enclave, and the response comes back out. Nothing sits on a server in readable form at any point.

The project uses confidential LLMs including Kimi K2.6, Gemma 4 31B, and GLM 5.1, all executed within these secure enclaves. Files and memory stored locally remain encrypted on the user’s own device rather than syncing to a backend database.

Integration with web services including Brave is part of the roadmap, pointing toward a browsing-aware assistant that handles sensitive queries without leaving a readable trail anywhere upstream.

Why Lido is building this now

Wisp traces its origins to late May 2026, when Lido identified an internal need for AI tooling that would not expose sensitive organizational data to third-party model providers.

The timing fits a broader strategic signal Lido sent in its May 2026 update, where the protocol explicitly flagged profitability and new product development as priorities.

No dedicated Wisp token has been announced. The project appears focused on product development and security validation rather than rushing to create a new asset.