- Brave Leo is the only major AI assistant that requires no account, stores no conversations, and proxies all queries through anonymous servers to strip identifying information.
- Free users access Llama, Qwen, Claude Haiku, and GLM models; premium subscribers ($14.99/month) get Claude Sonnet, DeepSeek R1, and higher rate limits.
- Leo works only inside the Brave browser on desktop and mobile, with no standalone app or third-party browser extension.
- The assistant can read and discuss the active browser tab, including web pages, PDFs, and Google Docs, but has no persistent memory across sessions.
What Happened
Brave Leo is an AI chat assistant built directly into the Brave browser, available on macOS, Windows, Linux, Android, and iOS. It launched with a privacy-first architecture that sets it apart from every competing AI chatbot. No account or login is required to use the free tier, conversations are never stored on Brave’s servers, and no data is used for model training.
As of 2026, the free tier provides access to Meta’s Llama 3.1, Alibaba’s Qwen 3, Anthropic’s Claude Haiku 3.5, and GLM models. Leo Premium, priced at $14.99 per month or $149.99 annually, unlocks Claude Sonnet 4, DeepSeek R1, and Kimi K2.5 with higher rate limits and priority access. A single premium subscription covers up to five devices. Brave now hosts all AI models, including Claude, through its own secure infrastructure.
Why It Matters
Most AI assistants require an account and retain conversation data for model improvement. ChatGPT, Gemini, and Copilot all collect usage data by default, requiring users to opt out through settings menus. Brave Leo takes the opposite approach: privacy is the default, and there is nothing to opt out of.
Queries are routed through a proxy server that strips IP addresses and other identifying information before reaching the AI model. Even premium subscribers remain unlinkable to their usage. Brave uses a credential-based token system for subscription verification that keeps payment information separated from chat activity. As of mid-2025, Brave hosts all AI models through its own infrastructure rather than routing queries to third-party providers, adding another layer of control over data handling.
Technical Details
Leo’s page-awareness feature reads the active browser tab in real time, allowing it to summarize web pages, analyze PDFs, and discuss Google Docs and Google Sheets content. Users can also use voice-to-text input on iOS. Unlike standalone chatbots that require copy-pasting content, Leo accesses page context natively through the browser, which eliminates the friction of switching between tabs and chat windows.
The Bring Your Own Model (BYOM) feature lets users connect local or remote AI models, or plug in third-party API keys for models not included in Brave’s lineup. This gives advanced users the ability to route queries to self-hosted models or specialized APIs while still benefiting from Leo’s privacy infrastructure.
Chat history is stored locally on the user’s device and can be deleted at any time. Brave sends only the current prompt and conversation context to its servers during a session. The company states that “Leo doesn’t retain or share chats, or use them for additional model training.”
Brave Leo also integrates with Brave Search to ground responses in current web information, reducing reliance on the model’s training data alone.
Who’s Affected
Leo is relevant to users who want AI assistance without data collection, particularly journalists, researchers, legal professionals, and anyone handling sensitive information. The free tier handles daily tasks competently for casual users. Premium appeals to power users who need frontier models like Claude Sonnet 4 without creating accounts with Anthropic or OpenAI directly.
The main limitation is browser lock-in. Leo works exclusively inside Brave. Users who prefer Chrome, Firefox, or Safari cannot access it. There is no standalone app, API, or browser extension for other platforms.
What’s Next
Leo has no persistent memory, meaning every conversation starts from scratch with no saved preferences or project-level context. For users who rely on ongoing projects or accumulated knowledge in tools like ChatGPT‘s memory feature, this is a significant limitation.
Brave has acknowledged that hallucinations remain “an intrinsic challenge in how LLMs work” and recommends users verify responses for accuracy. The company has not announced plans for persistent memory, cross-session context, or plugin integrations. Users evaluating Leo should weigh the privacy guarantees against the feature gaps compared to less private but more feature-rich alternatives.
