Browser automation

Most of my work happens through OAuth integrations — clean, fast, no surprises. But not everything has an API, and the gap shows up on exactly the workflows that drain your week. Government portals. Banking sites. That one niche tool your industry runs on. For those, I bring a real browser.

How it works

Underneath, I drive a headless Chromium with vision. I see what's on the page the way a human visitor does — rendered DOM, real layout, real screenshots — and reason about it: "click the next button", "fill the OTP into the second input", "find the download link in the table on row 3". Nothing assumes a specific HTML shape; if the page changes tomorrow, I adapt instead of breaking.

Each session is a fresh sandbox per user. No cookies carry over unless you explicitly told me to save a login. No browsing history is retained server-side. When the conversation ends, the session ends.

When the browser comes out

- Indian government portals — GST filing, RC renewal, FASTag recharge, PUC, passport status checks. These workflows are exactly the kind of 20-minute, 7-tab nightmare nobody enjoys.
- Banking sites — balance lookups, statement downloads, payment history. (Login flows with device binding may need a one-time push approval from you.)
- Niche industry tools that never published an API.
- Public web scraping — competitor pricing, public registries, schedules.
- Form filling at scale — registrations, applications, conference signups.
- Screenshots of confirmations and receipts you'll want later.

A real run

Guardrails

- I show you what I'm doing in real time — every navigation, every form fill — so you can stop me if something looks off.
- Anything destructive (paying money, submitting a final form, deleting data) prompts for your approval first. You can flip that to "always ask" or "only ask for amounts over ₹X".
- Every action is logged with a timestamp and a screenshot. Review later from the Records panel.

Limits

- Sites with aggressive anti-bot (banking login pages with hardware-token requirements, some captchas) may fail. When they do, I tell you exactly what blocked me — not a vague "I couldn't do it".
- Sessions cap at 10 minutes; longer workflows turn into scheduled jobs.
- I don't store screenshots beyond the conversation context unless you ask me to file them.

Pairs well with

- Records — when I OCR a renewed RC from the portal, the record updates automatically.
- Scheduled automation — turn one-time browser runs into recurring jobs (monthly GST check, weekly statement pull).
- Memory — I remember the login flow that worked last time, so the next run is faster.

Try this

> "Pull this month's GST filings from the GSTN portal and put a summary in Notion."

That's the kind of job the browser was built for.