A developer's retrospective on building a privacy-first browser history search extension

1. The Spark

Browser history search is broken. Every browser ships with a history page that can barely surface what you need. You type a keyword, get a chronological dump of URLs, and hope you recognize the right one. And if you want something smarter? The alternatives ship your browsing data to the cloud.

I wanted something different: a search engine that lives entirely on your machine, searches your history as fast as you can type, and never sends a single byte to any server. That became SmrutiCortex.

The name is Sanskrit — Smruti (memory) + Cortex (brain). A memory cortex for your browser.

2. The Build

Stack decisions were deliberate:

• TypeScript — Type safety from day one. No runtime type errors, no guessing.
• Chrome Manifest V3 — Future-proof from the start, even though MV3 was painful (ephemeral service workers, no persistent background pages).
• esbuild — Blazing fast builds (~5s dev, ~10s prod). Webpack would have been 10x slower.
• Vitest — Fast, ESM-native test runner that plays well with TypeScript and mocking Chrome APIs.

Architecture choices that shaped everything:

• 9 independent scorers — Each scoring dimension (title, URL, recency, visit count, etc.) is an isolated module. Change one without touching the rest. The search engine composes them.
• Service worker — Chrome MV3's biggest constraint. Ephemeral by design, so every state must be recoverable. IndexedDB for persistence, chrome.storage.local for settings, and a self-healing resilience module for graceful recovery.
• Shadow DOM overlay — The quick-search overlay (Ctrl+Shift+S) injects into any webpage without breaking its styles. Closed Shadow DOM means total CSS isolation.

The search algorithm — "Vivek Search": A graduated multi-token scoring system with intent-priority ranking. Not a simple substring match — it classifies matches into 4 tiers (EXACT, PREFIX, SUBSTRING, NONE), applies post-score boosters for strong multi-field matches, and ranks by coverage intent. The result: the page you're looking for is almost always in the top 3.

3. The AI Chapter