SkydiveAI coworkers that actually get work done
Promoted
Launching today
Tabbit AI
The best AI browser built both for you and your Agents.
247 followers
The best AI browser built both for you and your Agents.
247 followers
Tabbit is the AI browser that knows what you’re working on and can get the work done for you. Give it the pages, screenshots, and local files that matter, and Tabbit Agent can work across the web now or on a schedule. It delivers usable HTML, PDFs, and presentations, then saves the workflow as a Skill you can run again.
Tabbit AI
Hi Product Hunt! 👋
I'm Oliver, from the Tabbit team.
Quick backstory:
I spend most of my day in a browser. Whenever I'm deep in a project, everything I need is scattered across open tabs, local files, and screenshots I don't want to lose.
Then I'd open an AI tool and start copying everything over just so it knew what I was talking about. Once it gave me an answer, the clicking, checking, formatting, and exporting were still on me. There had to be a better way.
What makes Tabbit different:
Tabbit doesn't treat AI as a chat box bolted onto a browser. Point it at the pages, tabs, screenshots, selected text, and local files that matter, then give it a job to do.
Its GUI Agent can operate websites, run a task now or on a schedule, and create an HTML page, a PDF, or a deck you can use right away.
If you find a workflow worth keeping, save it as a Skill. Run it again, tweak it, or share it. We built Tabbit for people who build and research on the web—especially PMs, developers, and technical teams.
What we'd love your feedback on: What's the first tedious browser task you'd happily never do by hand again?
Thanks for taking a look. I'll be here in the comments and would love to hear what you think.
Tabbit AI
Hey Product Hunt! I’m Yu, the AI Product Manager at Tabbit.
Over the past year, I’ve focused on refining the browser experience and building our browser-use agent technology.
What I’m most proud of is the fast, token-efficient browser agent system we’ve built.
Across 75 runs using tasks from the BrowserBench benchmark, Tabbit achieved a 64% success rate. Compared with Agent Browser, it was approximately 1.9× faster while using 61% fewer input tokens.
Once Tabbit is installed, you can simply type `/tabbit` in agents like codex or claudecode to let it control the browser and work with your existing login sessions. It runs without taking over your browser, so you can continue using it while the agent works in parallel.
Tabbit is a browser that works beautifully for you—and smoothly for your agents.
I’d be happy to answer any technical questions about browser agents, Tabbit CLI, or how we built it! 🛠️
@liaocaoxuezhe Can Skills be shared with teammates, or are they mainly meant for personal workflows?
Tabbit AI
@david_turner12 Skills in Tabbit can be shared with your friends or published publicly in the Tabbit Skills Library.
For now, Tabbit don’t have team workspace features yet, so right now Skills can’t be shared across coworkers or organizations — only directly with friends through manual sharing.
Thanks for the feature suggestion!
whats in the 36% that fails? u published 64% over 75 runs on browserbench which most launches would just quietly skip
Tabbit AI
@niveditha_patluri1 Fair question — and honestly, the 36% mostly reflects current model limitations more than browser control limitations.
BrowserBench is useful precisely because it exposes where agents still break: long-horizon planning, ambiguous UI states, recovery from unexpected page changes, etc. We don’t think “hiding the misses” helps the industry progress.
What Tabbit focuses on is making browser interaction faster, more reliable, and more token-efficient for the model operating underneath. But if the underlying model cannot reason through a task reliably, no browser layer can magically turn that into 100% success.
We also expect completion rates to improve significantly with stronger next-gen models.
Side note: these runs were done using GPT-5.6 Luna, which is a relatively smaller model. Using larger models already improves the success rate noticeably — just with different latency/cost tradeoffs.
Tabbit AI
@olliez1 @ryee123123 You really know browsers — yep, Tabbit already does this.
Dodoboo
I just throw my tabs and screenshots at it and it gives me a deck back. No more copying everything into ChatGPT one by one. That alone saves me hours.
Awesome launch, folks! 👏
Tabbit AI
@saya_chen Thanks so much, Saya! Really glad the deck generation and Skills clicked for you.
RunEvr
@olliez1 @liaocaoxuezhe Downloaded Tabbit to give it a try! Congrats on the launch - really cool one!
Tabbit AI
@adana Thanks Adana! Really appreciate you giving Tabbit a try.
Tabbit AI
@charlie_titherley This is a good question.
In reality, during our operations, the tasks in Tabbit are diverse. I think the real significance of an Agent lies in freeing human hands and expanding the boundaries of human capabilities, enabling us to accomplish more tasks.
If you want an answer related to job types, I would say it's mainly divided into two categories: one involves tasks that go beyond the user's capabilities; the other includes tasks that can be efficiently and stably handled at a lower cost, replacing human operations.