Let’s talk
← All work

Project · On-device browser agent · 2026

Techie Mind

A privacy-first agent that uses the browser for you, in English or Kannada, typed or spoken. Local models first; every action passes a firewall and a verifier. Built for SIH26171 (ISRO).

Techie Mind, On-device browser agent

What it is

Techie Mind is a privacy-first AI agent that uses the browser for you. Tell it what you want in plain English or Kannada, typed or spoken, and it opens the site, searches, clicks, fills and reads, checking every step. It was built for Smart India Hackathon 2026, problem statement SIH26171 from ISRO: on-device visual perception for lightweight browser agents.

What it can do

  • Plain-language tasks like "search laptops under ₹50,000 on Flipkart", then "open the cheapest one".
  • 12 built-in skills, including summarise a page, deep research, compare prices, fill a form and monitor a page.
  • Kannada and voice input, translated on the device before anything runs.
  • Questions about an attached PDF, image or text file, answered by the local model.
  • Price monitoring that keeps running on a server while your laptop is off, with an email alert.

Built to be safe

Local models come first, so page content stays on your device. Every action passes an action firewall and a verifier before it counts. Payments, OTPs and CAPTCHAs always go back to you, and submitting a form needs your confirmation. It is open source under the MIT licence.

Speed and cost

Commands are understood in steps instead of sending everything to a big model. Plain code handles the simple ones; if it is unsure, a small local model called Laya decides in about 34 ms; only then does the larger Qwen 2.5-VL 7B model run, also locally through Ollama. Hard cases still get the big model, but most requests never need it, and running locally means no per-request bill.