Launching today
Murmure by Cyberesia Inc.
Private AI Transcription Assistant for Regulated Industries
5 followers
Private AI Transcription Assistant for Regulated Industries
5 followers
A private AI assistant for confidential transcription. Turn hours of speech into minutes of structured output. Record, import, query transcripts and files, export usable documents. Murmure is a native macOS app for confidential transcription, chat, and document generation, all local on Apple Silicon. Built in Geneva, Switzerland for government administration, lawyers, therapists, consultants, executives and more.
Hey Product Hunt 👋
I'm Kilian, founder of Cyberesia Inc. We just shipped Murmure, a Mac-native private AI assistant especially made for confidential recording, transcript and structured outputs.
The Why:
Every regulated professional I know was told "stop using ChatGPT for client work." Then "stop using cloud transcription app." Then "stop uploading anything." But the manual workflow (typing notes, hoping you remember) is brutal. There was no good third option.
We built Murmure as that third option. Native macOS, Apple Silicon only, everything local, including the AI.
Murmure runs open source models. Parakeet & Nemotron for transcription, MLX community models for generation, Ollama-compatible if you already have that set up.
What it does:
Three areas: Murmure (quick capture, live dictation in any app, chat over your transcripts), Studio (drop in long audio or video, pick a preset, get a structured document back), History (everything indexed and searchable, one place).
Where Murmure is genuinely different:
It's a document generation platform built on transcription, not only a dictation app.
- System-wide dictation in any macOS app (Mail, Word, Teams) with one shortcut. Try wiring that up yourself. Speak and it pastes the transcript into the active field. Entirely local.
- Dual transcription paths: Apple's fast on-device speech for quick capture, Parakeet for multi-speaker with speaker labels. On an M3 Pro, 1 hour of audio processes in ~5 minutes. M4 Max is faster. M1 base is usable but slower obviously.
- RAG over your own transcript library and indexed files, fully local
- 8+ document presets (legal notes, clinical drafts, meetings, podcasts...) with selectable sections, not a generic summarize button
- Reasoning / generation: Quantized 7B MLX-optimized. Ollama-compatible if you already have that set up. Used for chat, RAG over your indexed library, and structured doc generation with presets (meetings, legal notes, clinical drafts, podcasts, standups, courses, interviews) with selectable output sections (decisions, actions, risks, quotes, open questions).
- An optional MLX privacy filter that flags sensitive data locally before you share anything
- Output: Markdown, Word (.docx), PDF, all assembled locally.
- A guided first-launch auto-setup that finishes in under 5-10 minutes (model downloads).
Pricing: 189 CHF (~240 USD) one-time (Preview offer), two years of updates included. After the Preview window closes, the price moves to 289 CHF/year subscription (~367 USD) for new buyers. Preview buyers keep their one-time license forever
Any feedback is appreciated, especially from anyone who's tried to ship local-AI Mac apps before but I also have some questions :
- The preset system (meeting, workshop, clinical, podcast, interview, standup, course), what's missing from your workflow?
- Pricing is 189 CHF one-time for a two-year update window. Does that feel right for a pro tool in this category?
Mac with Apple Silicon (M1 or later) required. Happy to answer anything technical.
- Kilian