How are you planning to make the project sustainable (from a financial, and dev work/maintenance pov)?

plan is to sell licenses for Enterprise-version of browser, same as other open-source projects.

But we are much more performant than other libs (like playwright) which are written in JS, as we implement bunch of changes at chromium source code level -- for example, we are currently implementing a way to build enriched DOMtree required for agent interactions (click, input text, find element) directly at C++ level.

We also plan to expose those APIs to devs.

When someone in their infinite wisdom decides to refactor an api and deprecate the old one, it creates work for everyone downstream.

Maybe as an industry we can agree to do this every so often to keep the LLMs at bay for as long as possible. We can take a page out of the book of the maintainers of moviepy for shuffling their apis around, it definitely keeps everyone on their toes.

No invite system unlike bunch of others – you can download it today from our website or GitHub: https://github.com/browseros-ai/BrowserOS

--- Why bother building an alternative? We believe browsers will become the new operating systems, where we offload much bunch of our work to AI agents. But these agents will have access to all your sensitive data – emails, docs, on top of your browser history. Open-source, privacy-first alternatives need to exist.

We're not a search or ad company, so no weird incentives. Your data stays on your machine. You can use local LLMs with Ollama. We also support BYOK (bring your own keys), so no $200/month plans.

Another big difference vs Perplexity Comet: our agent runs locally in your browser (not on their server). You can actually watch it click around and do stuff, which is pretty cool! Short demo here: https://bit.ly/browserOS-demo

--- How we built? We patch Chromium's C++ source code with our changes, so we have the same security as Google Chrome. We also have an auto-updater for security patches and regular updates.

Working with Chromium's 15M lines of C++ has been another fun adventure that I'm writing a blog post on. Cursor/VSCode breaks at this scale, so we're back to using grep to find stuff and make changes. Claude code works surprisingly well too.

Building the binary takes ~3 hours on our M4 Max MacBook.

--- Next? We're just 2 people with a lot of work ahead (Firefox started with 3 hackers, history rhymes!). But we strongly believe that a privacy-first browser with local LLM support is more important than ever – since agents will have access to so much sensitive data.

Looking forward to any and all comments!

I don't have Mac or Windows.

still a team of 2 people, so bunch things on our plate.

But we strongly believe that for building a good agent co-pilot we need bunch of changes at Chromium C++ code level. For example, chromium has a accessibility tree for every website, but doesn't expose it as an API to chrome extension. Having access to accessibility tree would greatly improve agent execution.

We are also building bunch of changes in C++ for agents to interact with websites -- functions like click, elements with indexes. You can inject JS for doing this but it is 20-40X slower.

From Google's perspective, extension are meant to be lightweight applications, with restricted access.

See Sciter. A very cool, super lightweight alternative to Electron, but unfortunately it seems like a single developer project and I could never get any of the examples to run.

https://sciter.com/

What use-cases do you have in mind? like scraping?

Sciter uses quickjs and I just checked and its like 35-36x times slower than V8 JIT

Also another interesting rabbit hole is that I found Duktape in the quickjs benchmarks and I saw https://blogcpp.org/ as one of the projects within Duktape but I can't even see the project on github. We really need some better way of preserving open source stuff I guess

Not some rando's blog engine in C++, or other kinds of stupid throw-away code

Anyway <https://web.archive.org/web/20241122030659/https://github.co...> -> $(fossil clone https://code.rosaelefanten.org/blogcpp) which takes a stunning amount of time but then reports

    Round-trips: 15   Artifacts sent: 0  received: 2751
    Clone done, wire bytes sent: 4127  received: 124543126  remote: 2a03:4000:34:5e::1
    Rebuilding repository meta-data...
      100.0% complete...
    Extra delta compression... none found
    Vacuuming the database...
    project-id: 40a055cb170ae83c46b4ed9bf3b6a60e6e541aa0
    server-id:  cee9059305219c887fd29c677cbafb372252518a
    admin-user: mdaniel (password is "x2hEAaXUDj")
    opening the new ./blogcpp.fossil repository in directory ./blogcpp...
    3rdparty/ConfigParser/ConfigParser.cpp
    ...
    project-name: BlogC++
    repository:   /home/runner/blogcpp.fossil
    project-code: 40a055cb170ae83c46b4ed9bf3b6a60e6e541aa0
    checkout:     fbea390316bc3aace7de0a9ccdba90ecc1949a10 2024-12-17 22:03:16 UTC
    parent:       9e604d205e7922ef1af87952fe1bebef0cbac336 2022-01-15 02:22:33 UTC
    tags:         trunk
    comment:      fuck you github (user: Cthulhux)
    check-ins:    20

CSS2 is closer to trivial, but CSS3 is practically a 3D game engine with all of its matrix transforms, transitions, animations, variables - not to mention all the different layout schemes (Sciter blogged about introducing display:flex and display:grid two months ago)

The most interesting part of Sciter to me is that data persistence goes way beyond localStorage (string key: string value) or filesystem API, instead it's DyBase [0][1] behind the scenes, which looks to be a very intriguing style of storing trees of data in the host language's datatype (including whatever classes you define) without mucking about with the leaky abstractions of an ORM.

[0] https://github.com/c-smile/sciter-js-sdk/blob/main/docs/md/s...

[1] http://www.garret.ru/dybase.html

Can't imagine Firefox acquiring a Firefox fork!

I was tackling a similar problem few weeks ago and I found that playwright MCP was the most usable solution in my case. It doesn’t use an extension but it debugs the browser tabs (I guess using dev tools protocol) but I agree the experience was suboptimal

There are plenty of zero day exploit patches that Google immediately rolls out and not to mention all the other features that Google doesn't push to Chromium. I wouldn't trust a random open source project for my day-to-day browser.

Check out rtrvr.ai for a working implementation, we are an AI Web Agent browser extension that meets you where your workflows already are.

Chrome extensions is not a bad idea too. Just saying that owning the underlying source code has some strong advantages in the long term (being able to use C++ for a11y tree, DOM handling, etc -- which will be 20-40X faster than injecting JS using chrome extension).

Our benchmark results [https://www.rtrvr.ai/blog/web-bench-results] show that we are 7x faster than browser-use so curious to see if your claims live up to the hype

Given that you're working on a direct competitor, this comment reads as fearmongering, designed to drive people over to your product.

As an ex-Google engineer I know the immense engineering efforts and infrastructure setup to develop Chrome. It is very implausible that two people can handle all the effort to serve a secure browser with 15+ million lines of constantly changing C++ code.

A sandboxxed browser extension is the natural form factor for these agentic capabilities.

Definitely understand that keeping up with security patches is important. And this is an engineering challenge and not implausible to do -- Perplexity is 1/1000th the size of Google and they could be build a better product. So, "you can just do things".

We are still on day 1 of launch. We will only get better from here. And we won't be 2 people forever. We plan to hire, expand team and take on the engineering challenges.

I mean, I have no skin in the game but I mean, there are people who are using Dia (browser company) and Dia is closed source so it would be nice to see those people jumping to browser OS atleast.

I personally would prefer it as an extension but there are some limitations as the author of browserOS noted within extensions but I just wish that google/chromium can push those changes upstream I guess.

C++ APIs for dom tree, a11y. We eventually want to ship a small fine-tuned LLM and package with browser binary too.

Just getting started!

That's definitely a nice feature. Did you measure the impact on laptop battery life in a typical scenario (assuming there is such a scenario at this early stage)

If you run LLMs locally (using Ollama) and use that in our browser, that would impact battery life for sure.

Browser wars have begun.

> that OpenAI will be launching a (presumably not open source) browser of their own this summer.

For sure, won't be open-source. I bet in some parallel world, openAI would be non-profit and actually open-source AI :)

Kind of a sad browser war. The majority of these Chromium forks should exist, nor are they particularly viable. Sniffing out which one is going to be successful is obviously the hard problem.

Ideally we'd get an extensions API for AI agents and various companies could just release their own plugins. Sadly I don't think the majority is interested, they want to control the "experience".

What are the system requirements? And shouldn't they be listed on your website?

hardware requirements are minimal, same as Google Chrome, if you BYOK API keys for agents and are not running LLMs locally.

That's advantage of LLMs you dont have to go to the websites and soon you can just talk to your LLM like chatGPT or an advanced Siri to do this and boom it does all the research and shows you the top deals for you to purchase with a click (after you reviewed the prefered one and or talk to the LLM who makes the updates/edits you need before u do one click purchase).

If I was Orbitz I would be begging to work with Open AI / chatGPT (use Orbitz API) before Open AI does it themselves. Website visitation is and going to be dropping precipitiously along with the field of web design and development. That's my own field but it's so clear to me as technological evolutions that get adopted are the ones that reduce steps and make things easier to more magical.

You can bring your own API keys and change the default to any model you local.

Or better run model locally using Ollama and use that!

We are working on smaller, fine-tuned model too, which will be the default soon! It should be much faster and precise at navigation tasks.

So you rebuild your browser on every Chromium release? Because that's the risk: often changes go into Chromium with very innocent looking commit messages than are released from embargo 90 days later in their CVE reference

Yes, I'm acutely aware of exactly how much compute pulling off such a stunt requires; what I'm wondering is whether you are aware of exactly how much RCE risk you're running by squatting on someone else's C++ codebase that ships what feels like a vuln-a-week from one of the best funded security research teams in the world

It's also not a rando library that changes 3 times a year and who fucking cares if it has vulns: this is one of the biggest attack surfaces known to mankind given what it does for the user

Merely as a "for comparison," rebuilding Firefox takes about an hour on a developer class workstation, which IMHO sure would make tracking upstream a lot less expensive

I think if something can be done as an extension then there is no need to do it as a fork of the existing software.

Are there any differences b/w nanobrowser and brwoserOS? like some functionalities that only browserOS could do and not nanobrowser which are worth mentioning?

(Will you ever make a better FydeOS, or if you're laser-focused, perhaps be open to sharing some with them, so they could?)

I'll check out FydeOS!

I think it'd be better to show more non-trivial examples where the time savings is clear, and the failure cases are minimized... or even better how it's going to recover from those failure cases. Do I get a bespoke UI for the specific problem? Talk to it via chat?

This whole world is non-trivial. Good luck!

We are also just getting started and trying to narrow down on a high-value niche use-case.

There are few repetitive, boring use-cases where time saving could be meaningful -- one example: Walmart 3rd-party sellers routinely (multiple times a day) keep checking prices of the competitor products to price their products appropriately. This could be easily automated with current agentic browsers.

But for non-technical folks, agentic browsers seems like a good UX to build such and many more automations.

We were thinking of implement MCP protocol into the browser, so the browser can be an MCP server (that exposes bunch of tools -- navigation, click, extract) and you can connect that to your agent, would that work?

What is your use-case? Happy to chat on discord! https://discord.gg/YKwjt5vuKr

How does this make money? Surely this will have a cloud offering?

But if it doesn't make money, I can only assume that the team will be acqui-hired to answer that question.

Our plan is to make the consumer version of the browser for free and sell licenses for enterprise version.

This is similar to how many open-source projects sustain development.

And enterprise browsers are picking up -- Island browser, Talon browser.

Makes sense and the software license chosen (AGPL-3.0) also makes sense as well.

Sounds like a clear path to monetization and much better than some of the VC backed browser-first companies out there.

Good luck.

This is the real thing, the original if you will.

As for Perplexity, to me this company and line of product are seemed as the alternative to anything great in AI.

As a small startup we cannot outspend Perplexity Comet in marketing, so we just said open-source alternative so that people get what we are building.

From their GitHub readme:

> but Chrome hasn't evolved much in 10 years

Really?? It is not true. You guys please go and check release notes and commits log for Chrome/Chromium project for the past 10 years.

The main element in a web browser's UI is the web view where web pages get rendered. It may look the same 'rectangle' as it used to look 10 years ago. But the way chrome renders web pages and execute JavaScript have undergone a lot of changes over the years. Also they have added a lot of new standard and nonstandard HTML, CSS and JavaScript features. Then, there is WebGL 2.0, WebAssembly, WebGPU, etc.

Is Chrome an agentic browser at 35 million lines of code? That's what this project does. Whether they're successful at that or not is another story.

Since it's a Chromium fork, why not re-enable uBlock Origin instead?

The real question is, why not opt to fork Firefox who is doing that work for them.

But we are working on adding built-in adblockers just like Brave + enhancing it to detect more ad formats using lightweight local LLM.

So you're going to boot up my_llm.prompt("is this request an ad? %s" % request.copy_as_curl()) for every XHR on the page?!

I know this comment is on a page about an AI web browser, but let's turn down the "LLMs cure all problems" a notch

The jumpiness of pages switching and things changing when an AI is driving is extremely disorienting. I find it hard to follow a thread of continuity in the page flashes and ui changes as the bot acts.

Right now it’s like watching a screen recording with no hint as to what I’m “supposed” to be focusing on.

Regardless - I have use cases for this in the mcp/browser automation vein another user mentioned so super interested to see where this goes.

We will look into add a cursor movements; key typing should already appear like a human would (but probably we can slow it down a bit).

So far our patches have not conflicted with chromium updates. But there are viable ways to do patching and keep up with the pace (Brave Browser has pretty example infra setup to do this).

Also, I think it is super important we have different "webbrowser engines" and that are independent from big tech companies. If we only have chromium it would be really bad and hurt use consumer really bad and also stop inovation.

We really need to support independentCbrowsers like Firefox more.

You're privacy focused but use chrome as the engine?

We spoke to folks who had built browser on top of webkit and they spent nearly 2years just fixing random bugs and getting sites to work. I'm sure firefox/gecko engine would probably work better than webkit, but the point still is: if we don't use chromium, a lot of work does have go into fixing website compatibility issues, adding support for extension. We're 2 person startup and chromium codebase was easier to build on top of and provides a solid baseline.

And Brave has shown that you can build a privacy-focused browser on top of chromium still.

In the agentic browser era, I think there are so many low hanging fruits on privacy which are more important to address -- sending all your sensitive data to Perplexity Comet to sell ads is a pretty bad option right now. Supporting local LLM, letting folks bring their own API keys is crucial.

I tested it out asking it to summarize the comments of an Arstechnica article and at first it said:

> I am unable to summarize the comments as the provided page content only contains the article itself, not the comments section. The extract tool does not appear to be able to access the comments on this page.

I had to tell it to click on the "comments" link for it to actually start going through the comments. For reference, there's 3 pages of comments and it's taken more than 20 minutes and performed around 100 actions, with lots of those actions being scrolling down a very specific 1074 pixels. I am currently sitting here still waiting for the actual summary while it says "Validating task completion..."

Seems like it could be powerful, but the hand holding required and extremely slow speed make it unusable for me at this point.

Since I had Nxtscape installed, I tried the same experiment and it successfully did it in less time with fewer actions. Not sure if that's just random chance or if there's different logic running under the hood.

One final note: There's a Chrome extension that lets you use your iCloud passwords in Chrome and it doesn't work in either Nxtscape or BrowserOS. I'm not likely to use a browser that requires regularly having to go into my password manager to manually retrieve usernames and passwords, and I'm not changing my password manager for this.

Thank you for the feedback. Would love to chat more on discord (https://discord.gg/YKwjt5vuKr) and help you with your use-cases! we are shipping daily and improving things fast, agent should get much better in a few days.

> There's a Chrome extension that lets you use your iCloud passwords in Chrome and it doesn't work

Will check it out! We definitely want to make onboarding and password mgmt much easier!

Windows protected your PC Microsoft Defender SmartScreen prevented an unrecognized app from starting. Running this app might put your PC at risk.

a) they are real and b) they are credible.

Excited to use this

A lot of researchers who could really use this project would be sitting with a dual gpu setup on a linux machine. Would be an instant hit with the linux community.

I set the same simple task as I set Comet: examine the code resposible for choosing language captions and displaying them with a view to enhance the functionality to display multiple language captions in parallel. This is a high-school exercise, except for how atrocious the HTML can be for any given media-player page.

BrowserOS, like Comet, pretends it isn't capable of actually working with the DOM. This is because bad - or, at best, unambitious - design choices have been made as to how the LLM interfaces with the browser.

I even made a local HTTP API for Comet to use the MacOS accessibility API (which ironically gives greater scriptable browser control) and it couldn't even make a background HTTP get request with AppleScript to localhost. I suspect BrowserOS will be similarly crippled, although at least I can in theory fix it, unlike with Perplexity's closed-source half-assed Chromium bundling.

Can we try a little bit harder, dream a little bigger, imagine how to train, tune and instruct language modes to realistically fight the complexity of web bloat and give the user full programmatic natural-language power to interface with digital services?

I don't believe it would take any more effort than has already been employed.