Web persistent recorder
May 6, 2026Dashboard recorder with MediaRecorder Opus, IndexedDB persistence, offline upload via Service Worker, Wake Lock and recovery after crash.
Updated continuously. No vague 'Q3 2027' promises. If it's shipped, the date is here. If it's not, the status tells you why.
Persistence, EU residency, web/mobile capture. Product foundations.
Dashboard recorder with MediaRecorder Opus, IndexedDB persistence, offline upload via Service Worker, Wake Lock and recovery after crash.
AssemblyAI EU endpoint, Vertex AI in EU locations, R2 bucket with EU jurisdiction. Dedicated Hetzner server in Germany. Published sub-processors.
Client audio extraction with @ffmpeg/ffmpeg before upload (Opus mono 32kbps WebM). Auto-on for videos.
audiomap-whisper sidecar on Hetzner EPYC (faster-whisper 1.2.1, large-v3-turbo, int8 64 cores). Hybrid mode in production: AssemblyAI primary + automatic Whisper fallback.
Consent version 2.0 with the section “Responsibility for recordings” (STC 114/1984 and STS 22-04-2002) and a notice where you confirm the recording is lawful.
Flutter recorder optimization with interrupt handling (incoming call, app background, low battery). Pending tests with physical iOS/Android devices.
Multi-platform URL ingestion, agentic layer, Hispanic verticals.
13 templates × ES+EN (4 email, 5 doc, 4 slides). Generation via Vertex EU + Zod parsing + automatic refund on failure. Hispanic verticals included (tax, legal, medical, sales).
/print route with print-friendly CSS + browser native PDF. Zero deps, zero queue, zero Playwright pool.
It is switched on by percentage of accounts and by a list of test accounts, with a fixed split so each account always sees the same thing.
A 5-point rubric per deliverable type and template. Opening to 100% requires an average of at least 4.0 out of 5. It needs real recordings shared with permission to use them.
It opens first to a small group of accounts and is checked for 24 h. Then it goes to 10%, 50% and then every account within a week, watching for errors.
Paste a YouTube URL and it transcribes. Only YouTube + 1 landing (notasexperto.com) for 14 days. If conversion ≥ 10%, expand to IG/Spotify/TikTok/RSS.
Expansion after validating YouTube MVP. yt-dlp for IG/TikTok. Spotify Web API. Universal RSS parser for podcasts.
Already shipped on 2026-05-13. Pending: real quality eval + rollout.
First complete tax expert with benchmark + partnership. Sectoral verticals with their own domain authority.
/asesorias, /abogados, /clinicas, /consultoras, /coaches. Bilingual ES+EN, 10 SEO URLs.
fiscalexperto.com, notasexperto.com, audioexperto.com: DNS + Cloudflare + Next.js rewrite middleware.
BOE, AEAT publications, European tax regulations, TEAC doctrine. Base for fiscalexperto.
White-label revenue share or per-seat integration with Spanish tax platforms.
Public legal documentation and transparency.
/legal/aepd-compliance, /legal/dpia, /legal/tia, /legal/subprocessors, /legal/data-residency, /legal/comparison, /legal/request-deletion, /aepd.
Quick facts, compliance status grid (13 items), security practices, transparency commitments. ES+EN.
Internal draft ready. Pending: external counsel review + publication.
Quality-of-life improvements and depth. No categorical changes.
KPIs, distributions, top templates, rating breakdown, cost summary. For monitoring adoption + deciding GA.
/transcripcion, /diarizacion, /knowledge-graph, /entregables. Reusable FeatureLanding component. ES+EN.
Click on word → jump to timestamp in player. Favorites over transcript / ideas / analysis. Data already supports words with timestamps.
Audio + 12 key frames with client-side scene detection. Reduces uploaded data without losing visual context.
Persistent web recorder covers 90%. Defer until clear desktop demand signals.
Google Meet transcription without bot. Defer: the upload-first model covers the use case.
Next deliverables iteration. User uploads their system prompt for a template.
Persistent user voice profile. Applies to all templates. Reduces repeating context.
We do not promise dates we cannot deliver. But we do listen. If a feature is in "considering" status and you would pay for it, tell us: that moves it to "next up".
product@audiomap.ai