Tabductor
Browser agents that turn into fast, cheap scripts
Start new thread

Tabductor - Browser agents that turn into fast, cheap scripts

by•
Describe a browser task in plain English and watch an agent do it in a real browser. Tabductor turns that run into a script that reruns in seconds, with no model tokens. When the site changes, the agent steps back in and fixes it.

Add a comment

Replies

Best
Maker
📌

Hey Product Hunt 👋

I'm Ammar.

Browser agents are impressive the first time you watch one work. By the hundredth run, that feeling is gone 😅
They're slow, you pay for model tokens on every single run, and they never do things quite the same way twice.

I've spent a lot of time around compilers, and this looked like a familiar problem. A JIT doesn't reinterpret hot code forever. It watches it run, compiles a fast version, and puts guards on the assumptions it made. If a guard fails, it deoptimizes and falls back to the slow path.

Tabductor applies that idea to browser automation 🔄
The agent is the slow, flexible interpreter. The distilled script is the compiled code. Guards check that the page still looks the way it did during the recording. When a guard trips, Tabductor "deopts" back to the agent, which repairs the script.

⚙️ Here's how it works

  1. Describe the task in one sentence: the site, the inputs, and what you want back.

  2. Watch the agent do it live in a real hosted browser. If you need to log in or make a judgment call, grab the controls anytime 🕹️

  3. Tabductor distills the run into a static script with guards.

  4. Rerun it on a schedule, through the API, or over MCP, and it finishes in seconds ⚡ If the site changes and a guard trips, the agent steps back in and repairs the script for you 🔧

🎯 Where it shines: dashboards with no API, repetitive form filling, price and stock monitoring 📈, and getting data out from behind a login.


📦 What's included: signed-in browser profiles, per-workflow permission controls, residential proxies with country switching 🌍, CAPTCHA handling, a full recording and event timeline for every run, and typed results stored per workflow.


💰 Pricing: pay as you go, no plans.
• $0.02 per browser hour, billed by the second
• Model cost at provider price + 10%, charged only while the agent is actually running
• Distilled reruns use zero model tokens, so most runs cost next to nothing


I'd love to hear what you'd automate first, and what breaks 🙌