AI Tinkerers × OpenAI — Agents Everywhere, Cape Town

GuardianExtension

An agent that lives in your browser tab and blocks phishing before you hit submit.

scroll
01 / 08

Phishing doesn't happen
in a separate app.

It happens in the tab you're already using — a login page that looks right, a form asking for one detail too many. By the time something feels off, you've usually already typed it in.

02 / 08

The browser tab
is the environment.

A chatbot in another window can't see the page in front of you. GuardianExtension reads the actual page you're on, in real time, at the exact moment you're about to submit something.

03 / 08

How it works

1
Page loads
Extension reads the page in your active tab.
2
Agent is called
Context sent to the agent over A2A.
3
Agent reasons
Checks domain reputation and page content for phishing signals.
4
Exa grounds it
Answer API supplies real-time reputation data.
5
Verdict returns
Banner shows. Risky forms are blocked before submit.
04 / 08

It doesn't just warn.
It acts.

On a flagged page

A red banner appears with the reasoning and the specific signals found — brand impersonation, credential fields, urgency language. Submitting the form is blocked, with the choice to proceed anyway.

On a legitimate page

Nothing happens. No banner, no interruption. The agent only speaks up when there's something to say.

05 / 08

Built on

Go
agent runtime
Google ADK
agent framework & orchestration
A2A protocol
agent-to-agent & tool calls
Google Cloud Run
deployment
Exa
real-time grounding for the verdict
06 / 08

Built solo,
four hours.

The backend started as a template from a Google ADK + Go workshop the day before. Everything that makes this a phishing guardian — the check_url tool, Exa grounding, the CORS fix that let it talk to a real browser, and the extension itself — was built during the hackathon.

Sept 12, 2026 — AI Tinkerers × OpenAI, Cape Town.
07 / 08

GuardianExtension

Building in public. Building for Africa.
Built by Linford Musiyambodza — Linfy Tech Solutions
08 / 08