agent-browser is a browser automation CLI optimized for AI agents, built in Rust for speed. The current stable version is 0.26.0, with frequent releases (weekly/biweekly). It differentiates from Puppeteer/Playwright by being a native CLI with zero runtime dependencies (no Node.js required for the daemon), supporting headless Chrome via CDP, AI chat commands, and observability dashboard. It offers both accessibility-ref-based and traditional CSS selector interactions, and includes a daemon for persistent sessions.
npm install agent-browserVerified import paths — ran on the pinned version, not inferred.
Shows global install, Chrome download, navigating to a URL, taking a snapshot, clicking a ref, filling a form, getting text, taking a screenshot, and closing the browser.
Use execSync or spawn from child_process module to run commands.
Set AI_GATEWAY_API_KEY or use --key flag with the command.
Use --no-install or configure to use existing browser via CHROME_PATH environment variable.
Ensure scripts using --provider are updated to specify the provider name.
Run 'agent-browser stop' or kill the daemon process manually.
Upgrade to v0.23.4 or later.
Simply use 'agent-browser dashboard' without prior install.
Run 'npm install -g agent-browser' and ensure npm global bin directory is in PATH.
Run 'agent-browser install' or set the CHROME_PATH environment variable to an existing Chrome/Chromium executable.
Set the AI_GATEWAY_API_KEY environment variable or use --key with the command.
Run 'agent-browser daemon' to start the daemon, or use 'agent-browser open' to start automatically.