Murmur

A privacy-first AI meeting assistant that captures, transcribes, and summarizes meetings entirely on your machine. No cloud, no accounts, no data leaving your computer.

---

Murmur began as a fork of Meetily by Zackriya Solutions and has been heavily restructured since: the Python/FastAPI backend was removed entirely (everything runs inside the Tauri app now), the UI was redesigned, and several features were added on top. See What's different from upstream.

Introduction

Murmur runs entirely on your local machine. It captures your meetings (microphone + system audio), transcribes them in real time, and generates summaries — all locally. That makes it a good fit for anyone who needs to keep sensitive conversations under their own control.

• Privacy First: All processing happens locally on your device.
• Cost-Effective: Uses open-source AI models instead of expensive APIs.
• Flexible: Works offline and with any meeting platform (it captures system audio, not a bot in the call).
• Customizable: Self-hosted by definition — it's a desktop app you build and own.

What's different from upstream

• Backend-free architecture — the upstream Python/FastAPI backend and Docker stack are gone. Persistence is local SQLite inside the Rust core; summaries and chat run in-process.

• Redesigned UI — a dark-first interface with light/dark themes and a custom window chrome.

• Rust-native speaker diarization — identify who spoke, powered by sherpa-onnx, with mic masking and an optional speaker-count hint. No HuggingFace token or Python required.

• Transcript editor — edit segment text, merge/split segments, and reassign or rename speakers from the UI.

• Speaker-attributed AI — summaries and meeting chat see Speaker: text labels, so answers can attribute statements to people.

• Per-meeting chat — ask questions about a meeting, grounded in its transcript, with markdown rendering and message persistence.

• LM Studio provider — alongside Ollama, Claude, OpenAI, Groq, OpenRouter, custom OpenAI-compatible endpoints, and a bundled llama.cpp sidecar.

• Built-in MCP server — read-only Model Context Protocol access to your meetings database, built into the app binary. Point any MCP client (Claude Desktop, Claude Code, Cursor, …) at the Murmur executable with the --mcp flag — no Python, no separate install:

  {
    "mcpServers": {
      "murmur": {
        "command": "C:\\Program Files\\Murmur\\murmur.exe",
        "args": ["--mcp"]
      }
    }
  }

  (macOS: /Applications/Murmur.app/Contents/MacOS/murmur.) Tools: list_meetings, get_transcript, get_summary, get_meeting, search_transcripts. The database is opened read-only and the GUI does not need to be running. Override the DB location with --db <path> or MURMUR_DB_PATH if needed.

• Whisper output filtering — reduces YouTube-style hallucinations in transcripts.

• Markdown export for meetings.

• Zero telemetry — all analytics (PostHog) and the auto-updater have been removed. The app makes no usage-tracking or version-check calls; the only network connections are model downloads and any LLM provider you explicitly configure. See PRIVACY_POLICY.md.

Features

• Local First: All processing is done on your machine. No data ever leaves your computer.
• Real-time Transcription: Live transcript of your meeting as it happens, using Whisper or Parakeet models.
• Speaker Diarization: Identify and label individual speakers, locally.
• AI-Powered Summaries & Chat: Summarize meetings and chat with their transcripts using your choice of LLM provider.
• Professional Audio Mixing: Mic + system audio with RMS-based ducking and clipping prevention.
• GPU Accelerated: Metal/CoreML on macOS, CUDA or Vulkan on Windows/Linux.
• Multi-Platform: macOS, Windows, and Linux.

Installation

Build from source (see the Building guide for prerequisites and details):

git clone https://github.com/RenzoBeux/murmur
cd murmur/frontend
pnpm install
pnpm tauri:build          # auto-detects your GPU (CUDA/Metal/Vulkan/CPU)

Convenience wrappers with clean rebuild and logging:

• Windows: clean_run_windows.bat (dev) / clean_build_windows.bat (production)
• macOS/Linux: ./clean_run.sh (dev) / ./clean_build.sh (production)

System Architecture

Murmur is a single, self-contained application built with Tauri: a Rust core (audio pipeline, STT engines, SQLite, LLM clients) with a Next.js frontend.

For more details, see the Architecture documentation and CLAUDE.md.

For Developers

You'll need Rust, Node.js, and pnpm. For detailed build instructions — including GPU acceleration setup per platform — see the Building from Source guide and the GPU Acceleration guide. Contribution workflow is in CONTRIBUTING.md.

License

MIT License — see LICENSE.md. Copyright © 2026 Renzo Beux; original work copyright © 2024 Zackriya Solutions.

Acknowledgments

• Murmur is a fork of Meetily / meeting-minutes by Zackriya Solutions — thanks for open-sourcing an excellent foundation.
• Code was borrowed from Whisper.cpp, Screenpipe, and transcribe-rs.
• Thanks to NVIDIA for the Parakeet model, and to istupakov for the ONNX conversion.
• Speaker diarization is powered by sherpa-onnx.