Desktop app, web bookmarklet, multi-delivery architecture + full product audit #4032 — ccantynz/48co.v1

Originally written by @ccantynz-alt on GitHub.
Imported from https://github.com/ccantynz-alt/48co.v1/pull/5

Summary

Major product build: three delivery modes (desktop app, web bookmarklet, Chrome extension) plus a comprehensive audit that ensures every claim on the website matches what the code actually does. No false promises, no dead ends, no misleading features.

Architecture (new)

Desktop App (Electron) — System tray app with global hotkey. Whisper API + Web Speech API fallback. Types into ANY app via keyboard simulation.
Web Bookmarklet — Zero-download. Injects voice widget into any page. Voice commands, auto code fences, punctuation.
Chrome Extension — Always-on widget optimized for AI chat sites (Claude, ChatGPT, Gemini, DeepSeek).

Product audit fixes (latest commits)

Dishonest claims fixed:

Landing page demo was fake (just toggled a boolean) → now uses real Web Speech API with live transcript
"50+ languages" → corrected to "19+ / 50+" with context (web vs desktop)
"Free" pricing → now "Free*" with footnote about Whisper API cost (~$0.006/min)
"99%+ Accuracy" → now "High Accuracy" distinguishing Whisper vs Web Speech
Feature descriptions now specify which platform supports each feature

Broken code fixed:

Bookmarklet URL used location.origin which fetches from gmail.com when run on gmail → now bakes in correct 48co origin
Download page linked to non-existent .dmg/.exe files (404) → now links to GitHub releases with honest "Early Access" banner + build-from-source instructions
Desktop "Web Speech" engine just showed an error → added full Web Speech API fallback in overlay renderer
Desktop tray icon path referenced .png but only .svg exists → fixed
Voice commands and auto code fences were claimed but missing from bookmarklet → added

Other:

README completely rewritten (was referencing Faster-Whisper and Python backend)
Comparison table footnotes explain pricing and platform differences
All 8 ESLint errors fixed (unescaped entities)

Files changed

app/page.jsx — Live demo, honest claims, accurate comparison table
app/download/page.jsx — Early access banner, build-from-source, honest links
app/live/page.jsx — Fixed bookmarklet URL generation
app/install/page.jsx — ESLint fixes
app/layout.jsx — Updated meta description
app/globals.css — Fixed body overflow
public/inject.js — Added voice commands + auto code fences
desktop/ — Full Electron app with Web Speech fallback, fixed icon path
extension/ — Existing, maintained
CLAUDE.md — Engineering standards
README.md — Complete rewrite

Test plan

Landing page demo: click mic, speak, see transcript appear in widget
Bookmarklet: drag from /live to bookmarks bar, click on external site, verify widget loads from 48co.nz
Voice commands in bookmarklet: say "refactor this", verify prompt text inserted
Code fences: say something with coding keywords, verify markdown fence wrapping
Download page: verify links go to GitHub releases page (not direct 404 URLs)
Download page: verify "Early Access" banner is visible and honest
Desktop app: verify Web Speech engine works when selected in settings
All pages: npx next lint passes with 0 errors
All pages: npx next build succeeds (7/7 static pages)

https://claude.ai/code/session_01PBeNmPHrD2W6A2YKZRC7V3