Context switching
Close a tab, check downloads, search history — each action opens a new surface. You lose your place on the page you were actually working on.
Chrome extension
Manifest V3
Local-only
MIT
Your browser is an OS. This is its shell.
BrowserShell gives you a real terminal on top of any webpage — tabs, history, bookmarks, privacy tools, and page inspection as typed commands. Not another popup. Not another settings maze.
Press ` anywhere. Run tabs | grep github. Click row 2 to switch.
105 commands, pipes, aliases, and clickable output.
you@current-tab:~$
tabs | grep github
█
Install for Chrome
Build from source · Load unpacked at chrome://extensions · Press ` to toggle
105commands
0servers
0telemetry
`to toggle
Quake-style overlay · stays on the page you're working on
The problem
Your browser can do everything.
Getting it to do anything takes forever.
Chrome is an OS with dozens of subsystems — tabs, history, cookies, downloads, extensions, DevTools. But every task sends you through a different UI: menus, chrome:// pages, popups, and right-click detours.
Scattered tools
Tab managers, bookmark search, cookie viewers, and privacy cleaners live in separate extensions or buried chrome:// URLs. No single interface.
No composability
You can't pipe tab output into grep, alias your morning routine, or script "search history → open result → summarize page" as one flow.
Power users adapt
Developers already think in shells. The browser never gave them one — so they hack together shortcuts, bookmarks, and five extensions instead.
Why it was created
Built because the browser deserved a shell.
BrowserShell started from a simple frustration: the browser is the most-used app on your machine, yet the fastest way to triage tabs is still clicking through UI chrome designed for casual users.
Terminal users already have a mental model — commands, pipes, man pages, aliases. BrowserShell applies that model to browser primitives: tabs become listable rows, bookmarks become searchable paths, site data becomes commands you type and audit.
It's open source, runs entirely in your browser, stores config locally, and never phones home. If you live in the keyboard, this is the interface Chrome should have shipped.
Read the design philosophy →“Treat the browser like an OS. Give it bash.”
What it does
One overlay. Every browser capability as a command.
Press ` on any page. A fullscreen terminal drops in. Type. Click results. Close with Escape.
Your page stays underneath — no navigation, no new tab.
Command everything
Tabs, windows, bookmarks, history, downloads, cookies, DOM links, inputs, images, extensions — each exposed as a typed command with help and man docs.
tabs | grep docsbookmark search apidownloads
Compose workflows
Pipe output between commands. Set aliases for routines. Watch any command on an interval. Bang-expansion and a virtual filesystem for browser state.
history search react | headwatch 5 tabsalias ll=tabs
Click the output
Numbered rows are actions, not decoration. List tabs → click 3 to switch. List downloads → click 1 to open. A TUI that lives inside your browser.
tabs→ click rowlinks | head→ click hrefrecent→ restore session
Audit & forget
Inspect what a site stores, review permissions, and clear cookies or cache from the keyboard — with dry-run before destructive actions.
siteinfocookiesforget --dry-run
Before & after
Five clicks → one command
Without BrowserShell
- Right-click → "Search history for…"
chrome://downloadsin a new tab- DevTools → Application → Cookies
- Bookmark manager in another window
- Extension popup #7 for tab search
With BrowserShell
history search reactdownloads→ click row 1cookiesbookmark search docstabs | grep github
How it works
Three steps. You're productive in under a minute.
1Press
Press `
A Quake-style terminal overlays the current page. Works on any site. Toggle key is configurable in options.
2
Type a command
Tab completion, history, and Ctrl+R search. help and man tab when you're stuck.
3
Act on results
Click numbered rows to run follow-ups. Pipe output to grep, head, or the next command.
Try it
Live demos
Four real workflows. Switch tabs to explore — every command works in v1.0.0.
news.ycombinator.com
you@news.ycombinator.com:/$ tabs | grep -i hn
# Title URL
1 HN: AI agents thread news.ycombinator.com/item?id=…
2 Show HN: BrowserShell news.ycombinator.com/item?id=…
you@news.ycombinator.com:/$ tab switch 2
Switched to tab 2
you@news.ycombinator.com:/$ links | head
# Text Href
1 comments item?id=…
2 past news.ycombinator.com/past
you@news.ycombinator.com:/$█
you@tracker-heavy-site.com:/$ siteinfo
Cookies: 14 · Storage: 2.1 MB · Scripts: 23
you@tracker-heavy-site.com:/$ cookies | head
# Name Domain
1 _ga .tracker-heavy-site.com
2 session_id tracker-heavy-site.com
you@tracker-heavy-site.com:/$ forget --dry-run
Would remove: 14 cookies, 2.1 MB cache, 1.8 MB storage
you@tracker-heavy-site.com:/$ forget --preset strict
you@tracker-heavy-site.com:/$█
you@arxiv.org:/$ history search "transformer"
1 Attention Is All You Need arxiv.org/abs/1706.03762
you@arxiv.org:/$ go !1
Navigated to arxiv.org/abs/1706.03762
you@arxiv.org:/$ read | head -20
Abstract: The dominant sequence transduction models…
you@arxiv.org:/$ ai summarize --length short
Proposes self-attention architecture replacing RNNs…
you@arxiv.org:/$█
you@localhost:3000:/$ tech
React 18 · Next.js 14 · Vercel
you@localhost:3000:/$ reqs | grep -i api
1 GET /api/users 200 42ms
2 POST /api/auth 401 12ms
you@localhost:3000:/$ inputs
1 email input[type=email]
2 password input[type=password]
you@localhost:3000:/$ fill 1 "dev@example.com"
you@localhost:3000:/$█
Switch scenarios above — numbered rows are clickable in the real extension.
Command surface
105 commands. Nine categories.
Pipe them. Alias them. Watch them. All documented with man pages.
Navigation
go open back reload searchTabs
tabs tab pin discard domainWindows
windows recent sessions muteBookmarks
bookmarks bookmarkHistory
history search deletePage
links read click fill shot techPrivacy
forget siteinfo permissionsShell
ls cd cat grep alias exportUtility
options user config watch notifyVirtual filesystem
ls /tabs, cd /bookmarks, cat /current/url.txt — browser state as navigable paths.
Self-documenting
help, man tab, apropos privacy. Never stuck guessing syntax.
Local-only
No cloud. No telemetry. Config in chrome.storage.local. Audit the MIT source.
FAQ
Common questions
Why does BrowserShell need access to all websites?
The Quake-style overlay injects on every page so you can press ` anywhere. Page commands (links, read, click) run scripts in the active tab only when you execute them — never silently.
Does BrowserShell send data to any server?
No. Zero servers, zero analytics, zero accounts. Everything runs locally in Chrome. Config and history live in chrome.storage.local on your device.
How is this different from Vimium or Surfingkeys?
Those are keyboard navigation layers. BrowserShell is a full shell — pipes, VFS, 105 commands, clickable lists, aliases, watch mode. You manage tabs, history, bookmarks, downloads, and page DOM from one prompt.
How is this different from the Chrome DevTools console?
DevTools is for debugging JavaScript on one page. BrowserShell operates on browser-level resources (tabs, windows, bookmarks, history) and composes commands with pipes — closer to Bash than a REPL.
Can I use this without learning 105 commands?
Yes. Start with tabs, go, help, and man. Tab completion and search suggest commands as you type. apropos finds commands by keyword.
Is the forget command safe?
Always run forget --dry-run first. It shows exactly what would be deleted. Confirmation is required unless you pass --force. Named presets live in the options page.
Will this ship on the Chrome Web Store?
Planned. Privacy policy, permissions rationale, and security docs are ready. Until then, build from source and load unpacked from dist/.
Can websites extend BrowserShell?
On the roadmap. Sites will opt in via browser-shell.json manifests to register domain-specific commands when you visit them. See the Future page.
Open source
Read every line before you install
MIT licensed. No telemetry. Built in public.
Get started
Install in two minutes
Clone, build, load unpacked. Press ` on any page.
git clone https://github.com/jamalyusuf/browsershell.git
cd browsershell && npm install && npm run buildLoad dist/ at chrome://extensions → Developer mode → Load unpacked.