MindStack
Capture on select.
Generates in background.
Retrieve captured memory when Google search.

R - Reason

Every developer knows the feeling: you fix a bug, feel clever for a moment, and then two weeks later you’re staring at the same error, scrolling through the same Stack Overflow thread. That small cycle of rediscovery wastes more time than we admit, and it was the spark for MindStack. We wanted to build something that quietly remembers what you learn while you work and brings it back when you need it automatically. Not a note-taking tool, but a kind of memory: personal, automatic, and invisible.

E - Experience

When you highlight a solution online, a subtle “Capture” button appears beside your selection. One click saves a clean, tagged note - summarized and refined using Chrome’s on-device Gemini Nano APIs. Weeks later, when you search for a similar issue, MindStack quietly recognizes the match and surfaces a direct link to your own cheat sheet right beside the Google results. No setup. No cloud. Just Chrome that remembers.

C - Construction

MindStack is built with React, Vite, and TypeScript, packaged as a Manifest V3 Chrome extension. Captured text is processed through Gemini Nano’s Prompt API and Summarizer API to generate relevant tags and a short, clear explanation of the fix, then embedded locally using Xenova/all-MiniLM-L6-v2. Notes remain fully editable - you can refine or rewrite them anytime with the assistance of the Proofreader API. All data - vectors, notes, and metadata - lives in chrome.storage.local, enabling full offline recall. When you perform a Google search, the query is embedded, compared against local vectors, and the closest match is seamlessly injected into the results page with smooth React-driven animations. Everything happens client-side, instantly, and privately.

A - Artistry

Besides the technical challenges — what we cared most about was the subtlety of our extension. Making it feel native was everything. We didn’t want floating chatbots or “AI assistants.” We wanted Chrome itself to seem slightly more aware. That meant fine-tuning the “Capture" button design, handling asynchronous API calls without delay, and ensuring the recall card in Google Search feels natural. Most of the engineering went into restraint - building something that helps without announcing itself.

L - Lessons

Working fully client-side taught us that privacy, latency, and UX aren’t separate problems -they’re the same one. We learned to treat silence as a feature, and to trust the design that doesn’t draw attention to itself. When intelligence runs locally and fades into the flow of work, it stops feeling like AI and starts feeling like intuition.

L - Leverage

Next, we aim to make MindStack project-aware — grouping notes by repository and syncing (optionally, encrypted) with tools like Obsidian. The goal remains simple: a browser that remembers, quietly.