Brave Leo
Brave Leo is a browser-native AI assistant with no-account consumer access, local chat history, and multiple hosted or user-selected model options.
Model
Brave says its normal hosted route does not retain chats or use them for training, and it removes linkable identifiers such as an IP address from its AI-server workflow.
That is a meaningful privacy position. The hosted model runtime still sees the prompt while inferring. Feedback is a separate route: a user who submits feedback can send the conversation and optional page context for product improvement.
Brave has also announced a TEE-backed confidential-inference pilot in its Nightly channel. Its own description says Brave performs the first-stage verification and that end-to-end user-side verification is future work. Brave Leo TEE pilot preserves the distinction.
Lesson for the venture
Leo demonstrates three important ideas:
- browser-native local context can be useful
- model choice and local history are meaningful UX features
- privacy claims must distinguish normal no-logs hosting from confidential computing
The venture should be browser-neutral and should not require a customer to change their browser to get private AI. It can offer a native verifier, local document vault, and a privacy policy layer across browsers and devices.
Limit
Browser integration creates a broad context surface: page text, PDFs, documents, voice, feedback, and search can be attached to a request. Each needs its own route and consent boundary. Private AI routing treats tools and retrieval as first-class privacy risks, not a hidden implementation detail.