evrycallnote

Record every call. Transcribe, diarize, summarize — locally.

Cross-platform desktop app (Mac + Linux) that turns your calls into structured, searchable notes. 100% local AI processing. No cloud APIs, no subscriptions, no one listening.

Features

• One-click recording — system audio + mic, invisible to other participants
• Local transcription — whisper.cpp with GPU acceleration (Vulkan / Metal)
• Speaker diarization — who said what (pyannote.audio)
• AI summarization — key points, decisions, action items (Ollama)
• Obsidian export — markdown with YAML frontmatter, ready for your vault
• Remote processing — send audio to your GPU server over Tailscale
• Dashboard — browse, search, and replay past calls
• Offline-first — record without internet, sync later

Stack

Component	Technology
Desktop app	Tauri v2
GUI	Svelte 5 + TailwindCSS
Processing	Python 3.11+ / FastAPI
Database	Supabase (Postgres + Storage)
Transcription	whisper.cpp
Diarization	pyannote.audio
Summarization	Ollama

Quick Start

Prerequisites

• Rust (for Tauri)
• Node.js 20+ and pnpm
• Python 3.11+ and uv
• Ollama running locally
• whisper.cpp compiled with GPU support
• A Supabase project
• A HuggingFace account (for pyannote models)

Install

git clone https://github.com/m4zar/evrycallnote.git
cd evrycallnote

pnpm install
cd server && uv sync && cd ..

cargo tauri dev

Server Mode (optional)

Run the processing server on a GPU machine:

docker compose up -d
# Or: cd server && uvicorn main:app --host 0.0.0.0 --port 8000

Connect from the app via Tailscale IP.

How It Works

You click "Record"
  → captures system audio + mic
  → saves WAV to disk (streaming, crash-safe)

You click "Stop"
  → whisper.cpp transcribes (GPU-accelerated)
  → pyannote identifies speakers
  → Ollama summarizes
  → note saved to Supabase + exported to vault

~10-15 min processing for 1h call

Configuration

# ~/.evrycallnote/config.toml
[app]
mode = "local"              # local | remote
theme = "dark"

[server]
url = "http://100.x.x.x:8000"

[transcription]
model = "large-v3"
language = "auto"

[summarization]
engine = "ollama"
model = "llama3.1:8b"

[output]
vault_path = "~/Documents/vault"
note_dir = "research/meetings"

Secrets (Supabase keys, HuggingFace token) are stored in your OS keychain.

Output Example

---
title: "Weekly Sync"
date: 2026-03-30
duration: "42m"
speakers: 2
tags: [meeting, call]
---

## Summary
Discussion about portal v2 deployment...

## Decisions
- Prioritize auth before deploy

## Action Items
- [ ] Alex: configure auth — by Friday

## Transcript
[00:00] **Speaker 1:** Hi, let's align...
[00:15] **Speaker 2:** Sure, the timeline...

Architecture

evrycallnote/
├── src-tauri/           # Rust: Tauri core, audio, sidecar
├── src/                 # Svelte 5: GUI
├── server/              # Python: FastAPI processing engine
├── docker-compose.yml   # Server deployment
└── docs/                # Specification

License

MIT

Disclaimer

Recording calls may require consent from all participants depending on your jurisdiction. You are responsible for complying with applicable laws.