Do Browser
Do Browser is a Chrome extension that drives the browser in plain English. It scrapes pages, fills forms, exports CSV files and builds slides or small apps, for non-technical people who repeat the same web chores daily.
What is Do Browser?
Do Browser is a Chrome extension that turns natural-language instructions into real browser actions. Its own description is blunt: "Do Browser is an AI agent that controls Chrome with natural language. Describe what you want, watch it happen." There is no scripting layer and no visual workflow builder; the agent reads the page, decides what to click, type or copy, and works across several tabs at once.
The publisher is Smooth Brain LLC. The domain was first archived on 16 September 2024, the first public build, v1.3.0 "Early Access Launch", shipped on 6 December 2024, and version 3.0.0 left early access on 16 January 2026 after a full rewrite. The release log runs to 32 entries, the most recent being v3.1.33 on 21 July 2026.
Capabilities are grouped into four families: Automate (clicks, navigation, emails, forms), Create (interactive HTML slide decks), Build (small personal software with no code) and Analyze (data extraction, charts, summaries). The extension runs on Chrome only, on Mac, Windows and Linux desktops.
Three ways of supplying the model coexist. Paid plans include a hosted model. Alternatively you connect an existing ChatGPT or Claude subscription over OAuth, or paste your own Gemini API key; a personal key removes Do Browser's usage caps entirely and moves billing to the provider.
Two design choices shape the product. First, Skills: ten come pre-installed (Gmail, Google Sheets, Google Flights, Twitter/X, LinkedIn Messaging, Figma, tldraw, YouTube Transcripts, Slideshows, Quizzes) and follow the open Agent Skills standard, a folder holding a SKILL.md file, scripts, references and assets, so you can write your own or have the agent build one from a task it just completed. Second, a virtual filesystem backed by IndexedDB and split between /workspace for your files and a read-only /sys; the agent can run scripts and open files in a tab, and nothing is uploaded to a server.
The limits are stated by the publisher itself: hosted usage runs on a rolling five-hour window plus a weekly cap, no public API exists, and automated actions are supplied "as is", leaving the user to review and validate them.
What it does
- Drive Chrome in plain English, across several tabs, with no script and no workflow builder
- Extract data from any page and export it as a file, CSV included
- Fill in and submit web forms, then click, navigate and take screenshots as needed
- Summarize, analyze and chart the data found on a page
- Generate interactive HTML slide decks from existing content
- Build small personal apps such as to-do lists, trackers and dashboards, without code
- Learn reusable Skills, including ones created from a task the agent has just performed
When to use Do Browser / When not to
A quick filter to help you decide if Do Browser is the right fit.
When to use Do Browser
- Non-technical professionals stuck with repetitive browser chores: copy-pasting between tabs, pulling lists into a CSV, filling the same forms over and over
- People who already pay for ChatGPT Plus/Pro or Claude Pro/Max and can plug that subscription into the Free plan, adding nothing to their monthly bill
- Sales, recruiting and research profiles who need structured data out of live pages, such as exporting a LinkedIn connection list to CSV
- Knowledge workers running multi-tab routines: Gmail follow-ups, Slack digests, YouTube transcripts, page-referenced Q&A over long PDFs
- Designers and makers who want a Figma typography audit, an interactive slide deck or a small personal tracker built without writing a line of code
When not to use Do Browser
- Firefox, Safari or Edge users, and anyone working from a phone or tablet: this is a Chrome desktop extension for Mac, Windows and Linux only
- Engineering teams that need to industrialize automation outside the browser, since no public API is published and the docs hub holds four usage guides only
- Organizations with formal compliance requirements: no GDPR mention, no data processing agreement, no subprocessor list and no declared hosting country
- Anyone expecting unattended, guaranteed-correct execution, as the terms state that automated actions are provided "as is" and leave review and validation to the user
- Buyers who need an enterprise-grade vendor relationship: one contact address (the founder's), no published postal address and no company LinkedIn page
How to use Do Browser
A typical end-to-end flow, from setup to results.
- Install the Do Browser extension from the Chrome Web Store, on a Mac, Windows or Linux desktop
- Open the Do Browser side panel in Chrome
- Go to Settings > AI Provider and pick your engine: Do Browser (hosted), ChatGPT (sign in), Claude (sign in) or Gemini API Key
- For ChatGPT or Claude, complete the OAuth flow: you are redirected to the provider's sign-in page, then a green "Connected" indicator confirms the link
- For Gemini, paste an API key generated in Google AI Studio
- Describe the task in plain English; the agent picks the relevant Skill on its own, with no manual activation
- Drag and drop any documents the task needs into the file explorer (PDF, Word, spreadsheets); F2 renames a file, Delete removes it
- Collect the results in the Files tab under /workspace: preview, edit or download them, folders coming down as a zip
- Use Cmd+Shift+P for quick commands, voice input for dictation, and background mode to let the agent work without stealing focus
- Track consumption in Settings > Usage, which shows the rolling five-hour window and the weekly cap
Pros & Cons
Pros
- A genuinely usable free plan: connect the ChatGPT or Claude subscription you already pay for and add nothing to your bill, with no credit card required
- Nothing to configure, as the home page puts it: no setup, no scripting, no workflow builders, just one install from the Chrome Web Store
- Open choice of engine between the hosted model, an OpenAI or Anthropic subscription and a personal Gemini key, a personal API key removing the usage caps altogether
- Working material stays local: conversations on the device, virtual filesystem in IndexedDB, nothing uploaded to a server
- Extensible, interoperable Skills built on the open Agent Skills standard, and creatable straight from a conversation
- Sustained and documented release pace, with 32 versions listed and the most recent dated 21 July 2026
- Readable pricing at 25 and 100 USD per month, with usage caps honestly documented and counters visible inside the extension
Cons
- Chrome on desktop only: no Firefox, no Safari, no mobile app, and English is the sole declared interface language
- No public API, so nothing can be industrialized outside the browser
- Thin and outdated legal paperwork: privacy policy and terms are both dated 19 January 2025, while the product was rewritten in January 2026
- No mention of the GDPR, no Article 27 representative, no data processing agreement, no subprocessor list and no declared data hosting country
- The "not retained for training" statement is worded for the Max plan alone; nothing equivalent covers Free and Standard
- A very small publisher: a single contact address (the founder's), no company LinkedIn page, no published postal address
- Usage caps are never quantified beyond "generous usage limits", and buying extra credits is still announced as coming soon
Pricing & Plans
Do Browser operates on a freemium basis. A permanent free plan is available at no cost: you connect your own ChatGPT or Claude subscription, keep the core automation features and receive 10 free messages to try the hosted model, with no credit card required. The lowest paid entry point is USD 25.00 per month (Standard), which includes the hosted model with what the publisher calls generous usage limits. The Max plan costs USD 100.00 per month and provides four times the Standard allowance plus access to GPT-5.4. No annual pricing, per-seat rate or discount is published. Supplying your own OpenAI, Anthropic or OpenRouter API key removes Do Browser's caps and moves billing to that provider, while the purchase of additional credits is announced as coming soon.
- bring your own ChatGPT or Claude subscription
- keep the core automation features and get 10 free messages on the hosted model
- no credit card required
- hosted model included
- scraping
- natural language to automation
- file creation and export
- multi-tab workflows
- generous usage limits
- four times the Standard usage allowance and access to GPT-5.4
- Hosted usage on paid plans is metered on a rolling five-hour window plus a weekly cap
- the weekly cap being worth four times the five-hour allowance
Data, GDPR & hosting
A consolidated view of how Do Browser handles your data.
GDPR overview
There is no mention of the GDPR anywhere on the site: neither "GDPR" nor "General Data Protection Regulation" appears on any page collected, and no CCPA section exists either. No Article 27 EU representative, no data protection officer, no legal bases, no enumerated GDPR rights, no data processing agreement, no named subprocessors. The privacy policy of 19 January 2025, visibly built from a generic template, does grant a deletion right, exercisable from account settings or by email, and mentions international transfers only in general terms, naming no mechanism such as standard contractual clauses or an adequacy decision. The publisher is a US company whose policy defines its Country as California, United States, with no declared European presence. Privacy questions go to sawyer@dobrowser.io. On this evidence, neither compliance nor non-compliance can be stated.
Who owns the data?
Under the Terms of Service dated 19 January 2025, Smooth Brain LLC and its licensors own the extension, its software and its content; what you submit remains yours as User Content. Ownership is not the whole story: by submitting content you grant the publisher a non-exclusive, worldwide, royalty-free, sublicensable and transferable license to store, display, process and use it to provide and improve the Service. No end date is attached to that license. In practice much of your working material never leaves the machine: conversations are stored on the device and the virtual filesystem lives in the browser's IndexedDB, so clearing browser data destroys /workspace.
Reuse rights
The privacy policy of 19 January 2025 lists what is collected: email address, first and last name, and usage data such as IP address, browser type, pages visited and time spent. The stated purposes are broad: providing and maintaining the Service, managing your account, performing the contract, contacting you, sending commercial offers, and analytics. Sharing is broader still: service providers, affiliates, business partners "to offer You certain products, services or promotions", any merger or asset sale, and anyone you consent to. Session and persistent cookies, web beacons, tags and scripts are used. Third-party AI providers are involved, and the home-page FAQ states that "your data is not used to train their models". Your own outputs stay yours: the terms claim no ownership over the files and extractions the agent produces for you, but they do forbid automating anything that would breach a third-party site's own conditions, so responsibility for what you scrape and reuse rests with you.
Data retention & training
Hosting summary
No hosting country or region is declared anywhere on the site. The privacy policy of 19 January 2025 says only that personal data is processed "at the Company's operating offices and in any other places where the parties involved in the processing are located", and its definitions place the Company in California, United States. No data center, cloud provider, region or transfer mechanism is named. The marketing site itself resolves to 76.76.21.21, an anycast node on Amazon's AS16509 geolocated in the United States, but that describes the website's own infrastructure and is not a statement about where user data is stored. What is documented is local storage: conversations are kept on the device and the virtual filesystem lives in the browser's IndexedDB, with nothing uploaded to a server. Anything sent to a model does leave the browser, however: the terms state that Do Browser "leverages various AI models and technologies, including but not limited to Claude, OpenAI and Deepseek", while the policy mentions service providers only generically and publishes no list.
Things to keep in mind
Risks and trade-offs to weigh before adopting Do Browser.
- The privacy policy and the terms are both dated 19 January 2025 and were never revisited after the January 2026 rewrite, so the paperwork describes an older product than the one you install
- The privacy policy is visibly a generic template: it names Do Browser as the company while the terms name Smooth Brain LLC, and the words Smooth Brain appear nowhere in it
- The commitment that conversations are not retained for training is worded for the Max plan only, so do not assume it covers the Free and Standard plans
- The agent can type into password fields when you explicitly ask it to, and it works directly inside your own browser: treat sensitive accounts with care
- The terms forbid automating actions that would breach a third-party site's conditions, which places the legal risk of scraping on you rather than on the publisher
- Support, privacy and legal matters all run through one email address (the founder's), no postal address is published, the terms refer to the jurisdiction where Smooth Brain LLC is incorporated without ever naming it, and the domain holder is masked by an Icelandic privacy service
- Handing routine web work to an agent is convenient but can dull your own grip on the tools and the data involved; since usage caps are never quantified, plan for the day the agent stops mid-task and the job has to be finished by hand
Setup & Integrations
Technical difficulty
Very low. Installation is one click from the Chrome Web Store, and the home page claims no setup, no scripting and no workflow builders. On a paid plan the hosted model works out of the box with nothing to configure. Connecting your own engine means a five-step OAuth flow for ChatGPT or Claude, or pasting a Gemini key generated in Google AI Studio. Skills can be created by conversation, without learning the SKILL.md structure. The only documented snags are pop-up blockers that break the OAuth redirect and Google Cloud billing, which must be enabled for a Gemini key.
Deployment
Integrations
Supported languages
Behind Do Browser
Resources
All the official URLs gathered for verification and reference.
Frequently asked questions
Which browsers does Do Browser support?
Is there a free plan?
Can I use my own AI model?
Are there usage limits?
Is my data safe?
Can it access my passwords?
What can it automate?
Where do the files it produces go?
Are my files stored in the cloud?
Is there a referral program?
Should you pick Do Browser?
Do Browser sits in an unusual spot: a technically mature product carried by a very small structure. The code has clearly moved fast, with 32 releases, a full rewrite that left early access as v3.0.0 in January 2026 and updates still landing in July 2026, while the paperwork behind it has not moved at all since 19 January 2025.
What makes it worth a look is the economics. The Free plan is not a demo: you connect the ChatGPT or Claude subscription you already pay for and get the automation engine at no extra cost, plus 10 hosted messages to try the alternative. Paid tiers are legible, USD 25 and USD 100 a month, and a personal API key removes the caps entirely. The product asks nothing of you: no script, no workflow builder, one install from the Chrome Web Store. Working material stays on your machine, conversations on the device and files in the browser's IndexedDB, which is a real privacy advantage and a real fragility, since clearing browser data wipes it.
What should give you pause is everything around the software. There is no mention of the GDPR anywhere, no processing agreement, no subprocessor list, no stated hosting country, no postal address, and a single email address covering support, privacy and legal questions. The no-training statement is scoped to the Max plan alone, and the terms make you responsible for reviewing every automated action and for not automating anything that breaches a third-party site's rules.
The natural audience is individuals and small teams who want repetitive browser work done and can live with a one-person publisher. Organizations with compliance obligations, procurement checklists or an API-first integration plan should look elsewhere for now.
- Choosing a selection results in a full page refresh.
- Opens in a new window.