FAQ
Browsergent FAQ
Straight answers about the open-source AI browser agent for Chrome — setup, architecture, safety, and how it compares to classic automation.
What is Browsergent?
Browsergent is an open-source AI browser agent for Google Chrome. It lives in a side panel, observes the active web page, and completes plain-English tasks by generating JavaScript that runs through a sandboxed page.* command protocol.
Who is Browsergent for?
Developers, power users, and small teams who want Claude Code–style autonomy inside the browser: form filling, research, QA smoke paths, and authenticated internal tools — without standing up a remote browser farm.
How does the AI browser agent work?
You describe a goal. Browsergent snapshots the active tab into structured element refs. An LLM reasons over the task and snapshot, then returns JavaScript via a single tool called run_js. The extension runtime turns that into typed commands such as page.click, page.fill, page.goto, and page.snapshot, then feeds results back until the task ends.
How is Browsergent different from Playwright or Selenium?
Playwright and Selenium are scripted automation frameworks: you write selectors and control flow. Browsergent is an LLM agent. You state the outcome; the model chooses steps and recovers from common UI failures. It also runs on your real Chrome tab rather than a separate headless browser process.
How is Browsergent different from hosted browser agents?
Hosted products typically run in a vendor cloud browser. Browsergent is a local Chrome MV3 extension. Sessions, cookies, and keys stay under your control. You bring your own model API key; there is no Browsergent-operated model proxy required.
Is Browsergent free?
Yes. The project is dual-licensed MIT OR Apache-2.0. Download and use are free. You pay only your LLM provider under a bring-your-own-key (BYOK) model.
Which models and providers work?
Browsergent speaks the Anthropic Messages API. Use Anthropic directly, or compatible endpoints such as DeepSeek’s Anthropic-compatible base URL and z.ai / GLM. Configure API key, base URL, and model in the side-panel Settings.
Does Browsergent work on Firefox or Safari?
Not today. Browsergent targets Chrome Manifest V3 (side panel + content scripts). Ports to other browsers are not available yet.
Where does my API key go?
Your key is stored in the extension and sent only to the base URL you configure. Browsergent does not require a Browsergent cloud account for inference.
Can the agent access cookies and logged-in sessions?
Yes — by design it drives your real tab, so it can act inside authenticated pages the same way you can. That is powerful and risky. Review actions carefully and avoid high-stakes accounts while the project is experimental.
Is Browsergent safe for production accounts?
Treat it as experimental. The current philosophy exposes a wide capability surface so the project can learn browser-agent limits. Prefer low-risk profiles, watch the run, and stop the agent if it drifts.
What is run_js?
run_js is the model’s only browser tool. The LLM does not call chrome.* APIs directly. It writes JavaScript; the @pi-oxide/extension-js runtime executes allowed page.* operations and returns structured results or typed errors.
Can I install without building from source?
Yes. Download the prebuilt zip from the Download page, unzip it, enable Developer mode in chrome://extensions, and Load unpacked. You can also build from the GitHub repository if you prefer.
Where should I report bugs or request features?
Open an issue on GitHub: https://github.com/Irvingouj/Browsergent/issues