Back to vendors
B

Brave Leo AI Browsing

Also known as: Brave Leo, Leo AI, Brave AI Browsing

Visit site
Browser / computer-use agentindependentVerified 2026-07-22

Brave's privacy first browser agent: autonomous multi step browsing in an isolated profile with alignment checking and zero data retention, BYOM local model support, and the category's most documented security architecture; early access since May 2026.

Brave Leo is the privacy first AI in the Brave browser, and its AI Browsing mode (early access across all release channels since May 2026) gives it autonomous capabilities: actively browsing websites, researching across sources, filling shopping carts, comparing products, and completing multi step tasks while the user maintains control. Brave's safety architecture is the most documented in the consumer wave: the agent runs in an isolated browser profile with no access to cookies, logins, or sensitive data, alignment checking verifies intended actions, the mode is off by default, nothing is logged or retained, and Brave published the vulnerability research and responsible disclosures that defined the category's prompt injection risks, doubling its HackerOne bounties for agentic findings. BYOM lets users connect local models via Ollama or OpenAI compatible endpoints, with open models free and Claude and GPT class models in Premium; Skills provide reusable prompt shortcuts combinable with the agent. TEE backed deployments with NEAR.AI and Nvidia add cryptographically verifiable privacy.

Vendor details

Canonical URL

https://brave.com/leo-ai/

Category

Browser / computer-use agent

Funding status

Product of Brave Software; Leo launched 2023, agentic AI Browsing entered early access across all release channels May 2026

Company status

independent

Use cases & customers

Primary use cases

autonomous web researchproduct comparison and cart fillingpromo code discoverypage and document analysisprivate local model AI via BYOM

Target customers

privacy conscious usersresearchers and journalistsregulated industry professionalsconsumers

Deployment options

desktop (Windows/macOS/Linux)mobile (Android/iOS)BYOM local models

Integrations

Built into the Brave browser with page awareness across webpages, PDFs, Google Docs, Google Sheets, and YouTube; grounded by the independent Brave Search API; BYOM connects locally running models via Ollama or any OpenAI compatible endpoint.

Capability coverage

6.5 / 14 capabilities · 46%

Integrations & Tool CallingOperates through the browser UI with page awareness spanning webpages, PDFs, Google Docs, Sheets, and YouTube; no connector catalog or tool calling, Brave materials 2026-07-22 Partial
Workflow OrchestrationAI Browsing completes multi step tasks like cross site research, product comparison, and cart filling, but the capability is in early access and disabled by default, Brave materials 2026-07-22 Partial
Knowledge Grounding & RAGLive page awareness feeds active tab content as context without server upload, grounded by the independent Brave Search API index, Brave materials 2026-07-22 Partial
Human Oversight & GuardrailsAlignment checking verifies intended actions at runtime, users maintain control over agent actions, and the mode is optional and off by default, Brave help documentation 2026-07-22 Full
Security, Identity & GovernanceThe agent runs in an isolated profile with no access to cookies, logins, or sensitive data, with no logging or retention, TEE backed verifiable privacy via NEAR.AI and Nvidia, published vulnerability research with responsible disclosures, and doubled bug bounties for agentic findings, Brave materials 2026-07-22 Full
Observability & AuditabilityNo run visibility, action logs, or audit tooling documented; the no retention privacy design cuts against logging, Brave materials 2026-07-22 Unable to verify
Memory & State PersistenceSessions are deliberately not stored and inputs are not retained by design; persistence is explicitly against the privacy model, Brave materials 2026-07-22 Unable to verify
Deployment & Data ResidencyClient application across desktop and mobile platforms with a local only model option via BYOM keeping inference on device, Brave materials 2026-07-22 Partial
Prebuilt Agents, Templates & PacksSkills provide reusable prompt shortcuts combinable with the agent (such as /fact-check), short of a prebuilt agent library, Brave materials 2026-07-22 Partial
Triggers & Channel CoverageAll agentic activity is user initiated via direct requests or Skills shortcuts; no proactive or background trigger machinery documented, Brave materials 2026-07-22 Unable to verify
Model Flexibility & RoutingBYOM connects locally running models via Ollama or any OpenAI compatible endpoint, alongside bundled open models free and Claude and GPT class models in Premium, a documented full model choice surface including self hosted, TechRadar and Brave materials 2026-07-22 Full
APIs, SDKs & MCP ExtensibilityBYOM is model configuration rather than a developer API, SDK, or MCP surface; none documented, Brave materials 2026-07-22 Unable to verify
Testing, Debugging & OptimizationNo customer facing testing or evaluation tooling documented, Brave materials 2026-07-22 Unable to verify
Browser & Computer UseAutonomous browsing shipped to early access across all release channels: navigating sites, filling carts, and completing multi step tasks in the isolated agent profile, Brave materials 2026-07-22 Full

Pricing

Free tier + Premium

freemium subscription

Public — partialLow variable costFree tierTrial available

What is public

The freemium structure is public: Leo ships free in the Brave browser with open models and no account, and a Premium subscription adds frontier models. The Premium rate was not captured in retrieved materials.

Variable cost rationale

Free base tier and flat Premium subscription; BYOM local models remove even model costs for self hosted users, the lowest cost exposure in the consumer agent tier.

Sales call required

No — self-serve available

Free / trial

Free tier with no account required

Verified 2026-07-22

Contact us

Found a vendor we missed? Have feedback on the index? We'd love to hear from you.