Changelog

Plan changes

• Chat tools are now Pro/Trial — 🔎 transcript search, 📄 export transcripts (.txt) and 📋 whole-chat summary are now part of the Pro plan (and the 7-day Trial). On the Free plan, the buttons show a 🔒 lock with an invitation to check out Pro.
• Trial users keep using everything as before — nothing changes until the 7 days are up.

A tidier popup

• New 💬 Chat tools section — transcript search, daily digest and reminders became subsections inside a single accordion. The popup goes from 7 to 4 top-level sections: cleaner and easier to find things.
• Auto-send toggles together — the auto-send options for individual chats and for groups now sit side by side, in the same section.
• The link to configure keyboard shortcuts (chrome://extensions/shortcuts) is now always clickable on the popup home screen.

Improvements

• Productivity-focused daily digest — the daily summary got a dedicated prompt: organized per conversation (📌), highlighting action items and scheduled commitments. Less prose, more what-to-do.
• Click to copy — any item in the daily digest or search results goes to your clipboard with one click.
• Contact/chat name highlighted in search and digest results.

New

• 🗞️ Automatic daily digest (Pro/Trial) — at the time you choose, the AI gathers all the voice messages you received during the day and delivers a single summary, as a notification and in the ZapVox panel. No need to open chat after chat.

New

• 🔎 Search your transcribed voice messages — find any word ever said in your audios right from the side panel. Case and accent-insensitive, 100% local — nothing leaves your browser. Ctrl+F for your voice.

New

• 📖 Custom glossary (Pro) — teach the AI the terms of your daily routine (medication names, legal jargon, proper nouns) and transcriptions spell them exactly right. Saved to your account.
• E-mail sign-in — a third way to sign in: receive a code by e-mail, no Google or Microsoft account needed. The flow picks up where it left off even if the popup closes.
• Long audios transcribed end-to-end — multi-minute voice messages are no longer cut short: the transcription covers the whole message, start to finish.
• ⏱️ Time-saved counter in the panel — see how many minutes of audio you didn't have to listen to.

Fixes

• Phase 7 — extended retry for group audios — in busy groups, WhatsApp takes longer to hydrate audio metadata. Added an extra retry phase with 8s of additional wait after all other strategies fail, before showing a loading error.
• Friendlier audio error message — suggests clicking again manually before asking for an F5.
• Improved diagnostics in logs (includes isGroup and msgType flags).

Tests

• 397/397 tests green (4 new ones covering v3.4.14 error patterns).

Summaries in impersonal voice

• Summaries and topics no longer attribute actions to proper names — before: "Rafael needs to do X". Now: "X needs to be done". Avoids confusion (the reader can't tell if they are the sender/recipient/observer in a group) and eliminates attribution errors.
• Proper names are preserved when they truly matter ("Call Dr. Silva at 3pm", "PIX payment to XYZ Store").
• Edge Functions summarize v14 and topics v17 deployed.

Milestone

• ZapVox approved by Google and officially published on the Chrome Web Store. Available to everyone — install it free at chromewebstore.google.com.

Fixes

• Copy bug fix: duplicated watermark — copying a transcription manually duplicated the "🎙️ zapvoxia.com.br" watermark. Now built from a single deterministic source.
• CSS contrast fix — in dark mode, transcriptions could become unreadable due to a hardcoded color: #111. Fixed with WCAG AAA contrast in both themes.
• Manifest source consistency — Supabase host_permissions aligned between source and dist.

New

• Universal glass morphism on the transcription card — works with any WhatsApp theme and any custom wallpaper (photo, gradient, solid color). Near-opaque background with backdrop-filter: blur(14px) saturate(1.4) guarantees readability in 100% of scenarios.
• WCAG AA 7:1 contrast enforced.
• Automatic adaptation via prefers-color-scheme + body[data-theme].

New

• Sentiment cache — analysis survives an F5 (7-day TTL, no API cost on re-render).
• BETA badge in the popup.
• Adaptive theming via WhatsApp's native CSS variables.

UI cleanup

• Inline 🎭 Sentiment button — tone analysis right on the audio card. Result shown in a separate chip, never contaminating copy/auto-send.
• "Topics" button removed (orphaned since v2.2.1 merged topics into summaries).

Critical fix

• Bug: Portuguese audio transcribed in English. In "Automatic (SmartLang)" mode the Service Worker fell back to navigator.language, which returned en-US on Chromes with an English UI — forcing Whisper to transcribe Portuguese audio into odd English. Fixed to let Whisper auto-detect from the audio itself.

Fixes

• GET_USER_PLAN handler fixed — Pro users were blocked with "Scheduling is Pro-only" even while paying, because the handler was missing.
• Friendlier deleted-messages UX — instead of automatically replacing deleted content, shows an opt-in "🔓 View original (ZapVox)" button. More respectful and aligned with the CWS Single Purpose Policy.

Robust cascade

• Deterministic 5-phase cascade for audio loading (Phases 1a/1b/2/3/5) — replaced trial-and-error polling with WhatsApp's own internal primitives.
• Phase 5 — manual decrypt in the Service Worker — last resort when every other phase fails. The SW fetches from WhatsApp's CDN, derives keys via HKDF-SHA256, verifies HMAC and decrypts AES-256-CBC.
• Opt-in telemetry (client) — events sanitized through an explicit whitelist.

Structural changes

• Deleted message recovery (text + images + videos + audios) via local IndexedDB cache with a 7-day TTL. Opt-in in the popup.
• Sentiment analysis as a post-transcription action (BYOK + Cloud).
• Topics merged into Summary — summaries now automatically include the key points.
• Local Whisper removed — the product is now 100% cloud-first (BYOK Groq → BYOK OpenAI → Cloud Edge cascade).
• Free now includes 10 cloud transcriptions/day + unlimited BYOK.

Fixes

• Local transcription is now free and unlimited for all users
• The 3/day limit now applies only to cloud operations (summary, topics)
• Limit check moved to after local transcription in the transcription cascade

Security

• Removed an administrative debug handler — it allowed improper escalation to Pro
• Removed a debug handler — same vulnerability

Governance

• Version unified across all files (manifest, package.json, popup, docs)
• CHANGELOG rebuilt with complete v1.0–v1.7.7 history

Improvements

• Immutability of critical toggles protected
• web_accessible_resources consolidated in the manifest
• Stability improvements in audio detection

New

• Topic summaries — besides free-form summaries, now with key-point extraction
• topics Edge Function for server-side processing
• Summary/topics selector in the popup

Improvements

• BYOK Groq — your Groq key as Strategy 2 in the cascade
• AES-GCM 256-bit encryption for BYOK keys (PBKDF2 600k iterations)
• 7 exclusion layers + 5 positive strategies for audio detection
• Hardware compatibility page

Fixes

• PDFs/documents no longer mistaken for audio
• Native notification always active
• Translation auto-refresh token
• Duplicated transcription fixed

Launch

• Chrome MV3 extension for WhatsApp Web
• Local transcription with Whisper WASM
• Google login via Supabase Auth (PKCE)
• Free/Pro plans with Stripe
• 4-strategy transcription cascade
• 6 Edge Functions (auth, translate, summarize, topics, checkout, webhook)