browsertodo
Tell it what to do. Watch Chrome do it.
AI browser automation that runs in the Chrome you already use, signed in to your own accounts. Type a task, say it, or just press Enter on a page. It clicks, types and reads its way through, fast, and you see every step.
Chrome Web Store listing coming soon. Until then, install it from GitHub.
- Post today's tip from @alpha, @beta and @gamma
switch_x_account@alphaacttype the tip, click Post Jev 0.97 · 184 msswitch_x_account@betaacttype the tip, click Post Jev 0.95 · 179 msswitch_x_account@gammaacttype the tip, click Post Jev 0.96 · 191 msnavigatex.com/gamma/status/…- Posted on all three accounts, and each post is live on its profile.
- done
Ctrl+Shift+K on any page
One shortcut, three ways to start. On a Mac it's ⌘⇧K.
-
Type it
The side panel opens on this tab's conversation. Describe the job in plain words.
-
Say it
Press the shortcut again and talk; your words become the task. Voice input comes with paid plans.
-
Or say nothing
Press Enter with the box empty. It reads the screen and does the obvious next thing: a page that says “we sent you a verification email” gets your inbox opened and the link clicked.
Your real Chrome. Not a headless bot.
No separate automated browser, no copied cookies, no logging in again. It works in your own tabs, signed in to your own accounts, where you can see it.
- Already signed in
- Gmail, X, LinkedIn, your admin panels: if you're signed in, it can work there. Your cookies and passwords stay in your browser.
- In plain sight
- Its tabs sit in a tab group named “browsertodo”. Chrome shows its debugging bar while it works; close the bar and everything stops.
- It stops for you
- Login pages, 2FA, CAPTCHAs and account warnings pause the task and ask you. It never types your X password.
- One chat per tab
- Each tab keeps its own conversation, so two tabs can each run their own job at the same time.
Built to go fast
Most of an agent's time goes to finding things on the page. browsertodo cuts that down.
- Jev picks every click
- Claude describes the step, like “click Post”, and Jev, a small, fast model, picks the element in a fraction of a second. Claude sends several steps in one go.
- Many pages at once
- It opens up to 8 pages in parallel tabs and reads them in one step.
- Several jobs at once
- Tasks run side by side, each in its own tab. Tasks on X take turns, because your X accounts share one login.
- Streaming, start to finish
- You see what it's thinking and each step as it happens. Type to it mid-task, or press Stop.
5 pages, one by one4.8 s5 pages, parallel tabs0.5 sOpening and reading 5 test pages, measured.
Anything you do in a browser
Ask the way you'd ask a person. If you can do it in Chrome, it can.
- Social posting
- “Post today's tip from @alpha, @beta and @gamma.” It switches accounts, posts, then opens each post to check it's live.
- Inbox triage
- “Check my email and tell me what needs a reply.” You get a summary, sorted by what needs you.
- Sign-ups
- “Sign me up for the beta and click the confirmation link.” It finds the email in your inbox and clicks through.
- Research
- “Compare the pricing pages of these five tools.” It opens them side by side and reads them in one go.
- Forms
- “Register me for Thursday's meetup and stop before paying.”
Put it on repeat
For the jobs that come back every day, give a task a time and a repeat, and it runs without you. Part of every paid plan.
- Scheduled runs
- Three posts a day at 9, 1 and 6. An inbox summary every morning. Due tasks start on their own, with a short random pause between them.
- A TODO list in your account
- Add and check tasks from the side panel or the dashboard, with files attached.
- API keys
- Queue tasks from your own scripts and apps.
- Retries without doubles
- A temporary failure retries later, and checks first that the post didn't already go out. Three failures in a row pause the schedule and tell you why.
Bring your own Claude, or use ours
Point it at the Claude you already pay for, or sign in and go.
Open source
The extension and the local helper are MIT licensed. Read every line of what runs in your browser, build it yourself, or point it at your own server.
Read the code on GitHub- Web pages are treated as untrusted: it follows your instructions, not text it finds on a page.
- After it posts on X, it opens the post to check it's really there.
- Saved site logins are encrypted with your passphrase and never used for X.
- Every run is kept in the activity log, step by step.
Pricing
Free with your own Claude. Plans add usage credit for browsertodo AI, scheduling and the TODO list, API keys and voice.
-
Free
$0 a month
- Top up to use browsertodo AI in chat
- Not included: Scheduled runs and a TODO list in your account
- Not included: API keys to add tasks from your scripts
- Not included: Voice input
-
Starter
$9.99 a month
- $5 of usage credit each month
- Scheduled runs and a TODO list in your account
- API keys to add tasks from your scripts
- Voice input
-
Plus
$29.99 a month
- $20 of usage credit each month
- Scheduled runs and a TODO list in your account
- API keys to add tasks from your scripts
- Voice input
-
Pro
$199.99 a month
- $199.99 of usage credit each month
- Scheduled runs and a TODO list in your account
- API keys to add tasks from your scripts
- Voice input
Top up any time with $10, $25 or $50 of usage credit, on any plan. Top-ups never expire; monthly credit doesn't carry over.
browsertodo AI is charged from your usage credit at what the AI provider charges us, plus 30%. Your own Claude Code or API key costs browsertodo nothing, on every plan.
Prices in US dollars. Change or cancel your plan from the dashboard.
Give it something to do
Add the extension, press Ctrl+Shift+K on any page, and ask for something you'd rather not do yourself.
Chrome Web Store listing coming soon. Until then, install it from GitHub.