fastrepl/anarlog

anarlog – Privacy‑first AI meeting notepad

What it is – An open‑source, local‑first desktop (and mobile) app that records the audio from your own device, runs speech‑to‑text locally (or via a provider you choose), and lets you generate AI‑summaries, titles, or chat about the transcript. All data stays in a local SQLite database and plain files unless you explicitly enable the optional encrypted CloudSync.

Why it matters – Many AI meeting‑note services require a bot to join the call or send your audio to a cloud service. anarlog avoids that by capturing audio locally, offering on‑device transcription models on supported Macs, and letting you plug in any LLM you control (OpenAI, Ollama, LM Studio, etc.). This makes it suitable for security‑sensitive environments where data must not leave the machine.

Core features

Feature	How it works
Audio capture	Records directly from your microphone/device audio; no bot appears in the meeting.
Transcription	Uses on‑device models (Apple Speech, Soniqo) when available, otherwise any provider you configure (Soniox, OpenAI, custom STT).
AI summarisation & chat	Separate from transcription – you can point the app at Claude via OpenRouter, an OpenAI‑compatible server, Ollama, etc.
Local storage	Sessions, transcripts, and notes are saved in a local SQLite DB; recordings and attachments are plain files. Export to Markdown anytime.
Optional CloudSync	Encrypted sync/sharing is off by default; can be turned on via the built‑in Cloud service.
Cross‑platform	Desktop app built with Tauri v2 (Rust backend, React/TS UI) for macOS, Windows, Linux; mobile clients for iOS/Android; watchOS companion.
Extensible plugins	Tauri plugins expose STT, database, calendar, export, notifications, etc., and a CLI/MCP server can run skills (e.g., `skills/anarlog`).
Enterprise path	Source‑visible commercial components are provided for organizations that need a vetted, supported deployment.

Architecture snapshot

Frontend – React + TypeScript UI (desktop, web, mobile).
Backend – Rust crates for audio capture, diarisation, storage, and plugin glue.
Database – Local SQLite accessed via the Rust layer.
AI services – Configurable transcription provider + configurable LLM provider; UI shows the active model under Settings → Transcription and Settings → Intelligence.
Optional cloud stack – Supabase for auth, sync, billing; hosted API for AI, sharing, and integrations (in apps/api).

Typical workflow

Install – Download the binary for your OS (macOS, Windows, Linux) or build from source.
Join a meeting – Start recording; the app captures audio locally.
Transcribe – The selected STT model produces a transcript in real time or batch.
Summarise / chat – Press a button to ask the connected LLM to generate a summary, title, or answer questions about the transcript.
Edit & export – Treat the generated note like a document, edit, and export to Markdown.
(Optional) Enable encrypted CloudSync to share notes or sync across devices.

Getting started (from the repo)

# Prereqs
brew install process-compose   # macOS example
# or run scripts/setup-linux.sh on Debian/Ubuntu

pnpm install --frozen-lockfile
pnpm dev:desktop   # launch the desktop app in development mode
# pnpm dev:web      # run the website front‑end

The repo includes a CONTRIBUTING.md with details on configuring optional back‑ends, running the Supabase stack, and building the mobile/watch apps.

Who should use it?

Security‑focused teams that cannot send audio to third‑party clouds.
Developers / power users who want a fully auditable, MIT‑licensed meeting‑note app they can fork or self‑host.
Organizations looking for a path to an enterprise‑grade, source‑visible deployment.
Anyone who wants AI‑generated meeting summaries but prefers to keep the raw data on their own machine.

License & community

Community code – MIT license (free to fork, audit, commercialize).
Enterprise components – Commercial license (source‑visible but not MIT).
Community – Discord, Reddit, X, and a public status page; contributions are welcomed via pull requests.

Bottom line – anarlog is a genuine, privacy‑first AI meeting‑note application that lets you control both the transcription and the language‑model stages, with a fully local data store and optional secure sync. It sits squarely in the AI‑productivity space and is built with modern open‑source tools (Rust, Tauri, React, TypeScript).