Browser automation CLI built for AI agents Browser-act, a new CLI tool designed for AI agents, enables automated browser control with anti-bot bypass, human handoff, and parallel task isolation. The open-source tool supports stealth fingerprinting, CAPTCHA solving, and token-efficient text output for LLM reasoning, with free basic features and paid options for managed proxies and additional stealth browsers. Browser automation CLI built for AI agents. Get past anti-bot walls, hand off to humans across platforms when stuck, run parallel tasks without cross-contamination, and isolate multiple accounts in independent browsers. The browser an AI agent needs has to reach places standard tools can't, let a human seamlessly take over when the agent is stuck, keep parallel tasks from cross-contaminating, and be designed for LLM reasoning — not human-written scripts. A browser for agents must get four things right. 1. Break through blocks — three progressive layers Environment layer — stealth fingerprint spoofing, TLS rotation, proxy switching. The vast majority of blocks never trigger. Execution layer — solve-captcha auto-solves CAPTCHAs; stealth-extract pulls protected pages in one command. Human layer — remote-assist generates a live URL; the user takes over from any device, and the agent continues seamlessly when done. 2. Three browser modes — by real-world scenario | Mode | Scenario | Key trait | |---|---|---| chrome | Reuse local Chrome login state | Profile import or CDP attach | stealth privacy mode | Frictionless batch scraping without login | Fresh fingerprint per session + proxy rotation, zero residue | stealth fixed identity | Logged-in accounts · multi-browser parallel | Stable fingerprint + stable IP, stable account identity, not flagged as bots | 3. Zero-interference concurrency — every agent in its own lane - Cross-browser parallel — independent cookies, fingerprints, proxies. Sites cannot correlate them. - Same-browser multi-session — shared login state, independent execution, tasks don't block each other. - Privacy mode — fresh fingerprint and empty profile per session, zero residue when done. 4. Designed for agent reasoning — not human scripts Compact text output — indexed text format, several times more token-efficient than JSON or HTML. Indexed interaction — state returns an indexed list; click 3 / input 2 "..." . No DOM parsing required. Semantic memory — every browser carries a desc , matched to tasks by meaning. Concurrency-safe — session ownership + explicit naming. Multi-agent operation never conflicts. Security: confirmation gating — sensitive operations browser create / delete, Profile import, proxy changes, security and privacy toggles require explicit user approval. Prior approvals do not carry over. Enforced at the Skill layer, not a configuration toggle. Better headless — Default headless without disrupting users; stealth headless that isn't detected. Cross-platform remote handoff — Any device opens the link to take over, and the agent continues seamlessly. Tell your AI agent: Install browser-act. Skill source: https://github.com/browser-act/skills/tree/main/browser-act . Verify it works after installation. Extract protected page content zero config browser-act stealth-extract https://example.com Full browser automation browser-act --session my-task browser open