Tandem Browser offers a novel approach to web browsing by enabling symbiotic interaction between humans and AI agents. It functions as a local-first Electron browser where an AI agent operates alongside the user, accessing tabs, pages, and authenticated sessions. This allows for powerful AI-driven automation and assistance directly within the user's existing browser context, benefiting developers, researchers, and power users seeking advanced AI integration without complex scraping or API management.

How It Works

Tandem Browser facilitates human-AI symbiosis through a local-first Electron application. An AI agent integrates directly into the user's browsing session, mirroring their activity and operating within their authenticated contexts. AI interaction is managed via the Multi-modal Communication Protocol (MCP) or a comprehensive 300+ endpoint HTTP API, ensuring compatibility with various AI clients like Ollama or Cursor. A key architectural differentiator is its 8-layer security model, designed to protect the AI agent from web content threats and prompt injection, while maintaining the illusion of a standard browser to websites.

Quick Start & Requirements

• Primary install/run command: Clone the repository (git clone https://github.com/hydro13/tandem-browser.git), navigate into the directory (cd tandem-browser), install dependencies (npm install), and start the application (npm start).
• Prerequisites: Node.js (implied by npm).
• Platform: macOS is the primary supported platform; Linux is functional. Windows has not been validated.
• AI Connection: Connect via MCP clients (e.g., Claude Code, Ollama) or the HTTP API.
• Links: GitHub Discussions for community Q&A and ideas.

Highlighted Details

• Extensive Tooling: Provides 239 distinct tools for AI agents, covering navigation, tab management, page content extraction, accessibility, DevTools, network inspection, session management, bookmarks, password handling, extension control, workflow automation, live previews, media interaction, and device emulation.
• Robust Security Model: Features an 8-layer security architecture, including network shields, outbound data scanning, JavaScript analysis, behavior monitoring, prompt injection defense, and human-in-the-loop escalation for risky actions.
• Shared Context & Stealth: The AI agent perceives the user's browsing activity, and vice-versa. Websites interact with Tandem as a normal Chrome browser, unaware of the AI agent's presence.
• Local-First Architecture: All features, including sidebar applications (Telegram, Discord, etc.), workspaces, pinboards, password management, and Chrome extension support, operate locally without cloud dependency.

Maintenance & Community

Tandem Browser is currently in a public developer preview state, actively seeking contributors. Community interaction and support are primarily channeled through GitHub Discussions for Q&A, feature ideas, and showcasing setups. Specific bug reports and feature requests should be filed as GitHub Issues. Key contribution areas include MCP tool development, browser API enhancements, Linux support, and security hardening.

Licensing & Compatibility

The project is licensed under the MIT license. This permissive license generally allows for commercial use and integration into closed-source projects without significant restrictions.

Limitations & Caveats

As a public developer preview, Tandem Browser is not yet a polished, mass-user release. Its primary development focus and validation are on macOS, with Linux support being functional but secondary. Windows compatibility is explicitly unvalidated. Binaries are not yet published, requiring users to build from source.