# I Built Termsight: An AI That Reads the Privacy Policy So You Don't Have To.

> Source: <https://dev.to/ayomiposi20/i-built-termsight-an-ai-that-reads-the-privacy-policy-so-you-dont-have-to-7pa>
> Published: 2026-10-25 16:09:50+00:00

What I Built

Termsight is an AI tool that reads privacy policies and terms of service

and explains them in plain English, so you don't have to.

Paste in a policy (or a link to one), and Termsight breaks it down into

four things people actually care about: what information is collected,

how it may be shared with third parties, how long it's retained, and

what rights you might be giving up. Every finding links back to the

exact section of the original document it came from — and if the

policy doesn't clearly answer something, Termsight says so instead of

guessing.

The lazy payoff: nobody has to actually read pages of legal text

anymore just to know what they're agreeing to. Anyone who's ever

clicked "Accept" without reading gets to be a little lazier and

better informed at the same time.

[https://lumping-playable-sauciness.ngrok-free.dev/](https://lumping-playable-sauciness.ngrok-free.dev/)

GitHub: [https://github.com/Ayomiposi20/termsight](https://github.com/Ayomiposi20/termsight)

Termsight runs on Qwen2.5-1.5B-Instruct, an open-weight language

model, loaded and run locally rather than called through a commercial

AI API. The interface is built with Streamlit.

The pipeline:

I deliberately split the analysis into four separate, focused prompts

rather than one big prompt asking everything at once with a smaller

model, this made a real difference in answer quality and prevented the

model from just dumping bare citations instead of actual explanations.

Privacy policies are sensitive by nature pasting one into a

commercial AI service means trusting yet another third party with that

text, which is a strange thing to do for a tool whose whole purpose is

helping you protect your privacy.

Because Termsight runs on an open-weight model instead of a closed API,

the policy text is never sent to a third-party AI service. That's not

just a technical detail it's the actual point. A tool meant to help

you understand what you're agreeing to shouldn't require agreeing to

something new (sending your data elsewhere) just to use it. Open models

make that kind of honest, self-contained tool possible in a way a

closed API fundamentally can't.