An agent for the browser you already use.
Read signed-in pages, take action, create files, and follow up locally.
English · 简体中文
Quick start · Configuration · Documentation · Current limits · Contribute · Issues
Browser Kitten is an MIT-licensed Chrome MV3 extension with an agent in the side panel. It works with your existing browser session and keeps conversations, workspace files, and follow-up tasks on this computer. Model and other provider requests use the services you configure.
Current source version: 1.2.0 · Load unpacked · No build step for installation.
| Capability | In practice |
|---|---|
| Read & research | Understand pages you are signed in to, select relevant content, and collect findings. |
| Act on pages | Delegate clicks, form entry, and navigation to a separately configured execution model. |
| Create deliverables | Save documents, HTML pages, data, and files in a session workspace. |
| Follow up | Arrange local checks, preserve progress, and manage ongoing tasks from a collapsible drawer. |
See the actual side panel — light & dark
Captured from the extension in a fresh browser profile. The example is an unsent draft, not a completed run.
- Download the current source ZIP and extract it into a folder you will keep.
- Open
chrome://extensionsin desktop Chrome, enable Developer mode, then choose Load unpacked. SelectBrowserKitten-main, the folder that directly containsmanifest.json. - Pin Browser Kitten, refresh an ordinary website tab, open the side panel, and configure your providers below.
You do not need Node.js, Python, npm, or a build command to install it. The source folder itself is the extension load root.
Trying the latest work? The download above follows
main. The separate GitHub Releases page may contain older tagged versions.
For the full walkthrough, permissions, update instructions, and troubleshooting, see Getting started · 中文安装指南.
Open Settings from the side-panel menu. For a bring-your-own-key setup:
| Setting | Needed for | Enter |
|---|---|---|
| Inference | Planning, analysis, and tool calls | Base URL, API key, and a tool-capable model ID |
| System1 | Page clicks and form interactions | A separate execution-model configuration using an OpenAI-compatible /chat/completions endpoint |
| Image / Transcription / Search | Additional media and web capabilities | Configure only the services you need |
Inference and System1 can use the same provider, but must be configured separately. The unpacked build accepts your own keys without a Browser Kitten subscription; the external providers still charge for usage. Account-hosted services are optional and depend on the separate backend.
Try a small first task:
Summarize the current page and save its key findings as a document.
Enter inserts a newline. Ctrl+Enter / Cmd+Enter sends. Enable page scripts and capture permissions only when the task needs them; see the permission guide.
Describe the goal, permitted actions, reporting conditions, and when to stop. The agent can arrange or remove wakeups within that goal. Manage them in the collapsible task drawer above the input.
- Chrome must remain running and the computer awake. Closing the side panel is fine.
- Checks run on a minute-scale schedule; immediate webpage-event subscriptions are not implemented.
- New scheduled tasks can stay quiet when nothing meaningful changes, and report actions, changes, or requests for help.
See follow-up behavior and controls.
- This is a local browser extension; there is no cloud execution environment or native OS / terminal control.
- The host currently enforces per-tab leases and per-session execution slots, rather than a global browser lock. Automatic grouping of all agent-controlled website tabs is not implemented. Run one browser-operating task at a time.
- Restricted Chrome pages and the Web Store are not ordinary automation targets. Browser smoke tests do not establish success on every website or with every model.
- Settings and workspaces are local to this Chrome profile. They do not automatically move to another computer. Relevant task content may be sent to your configured providers.
Read updates and data before replacing an installation.
| Start here | Go deeper |
|---|---|
| Getting started · 安装指南 | Architecture · Agent runtime (中文) |
| Documentation index | Tools & Workflow · Workflow contract (中文) |
| Troubleshooting & support | UI design · Media (中文) |
| Changelog | Security policy · Third-party notices |
Bug reports, documentation fixes, and focused pull requests are welcome. Start with CONTRIBUTING.md for setup and checks. Use a bug report or feature request to give maintainers useful context.
Please follow the Code of Conduct. Report vulnerabilities through private security reporting, rather than a public issue. No response-time commitment is implied.
MIT © 2026 Yiteng (Player-YN). Built with open-source projects including Vercel AI SDK, QuickJS, esbuild, Zod, Lucide, and fflate. See third-party notices for their respective licenses.