What it does

anarlog sits in your meetings, records audio, transcribes it locally, and spits out notes as plain Markdown files on disk. You bring your own LLM—OpenAI, Anthropic, Ollama, whatever speaks OpenAI-compatible APIs—and it handles the summarization. No accounts, no hosted backend, no telemetry. Built with Rust, Tauri, React, and TypeScript.

The interesting bit

The “local-first” pitch is usually hand-wavy, but here it’s structural: transcription runs on-device, audio never leaves, and every meeting becomes a .md file you can grep, git-track, or Syncthing-sync. The team has since moved on to a commercial product called char, but anarlog stays maintained as the open-source forkable path.

Key highlights

• On-device transcription; audio stays local
• Markdown-on-disk output with no proprietary format lock-in
• BYO LLM: works with OpenAI, Anthropic, Gemini, OpenRouter, Ollama, LM Studio, or anything OpenAI-compatible
• MIT license: fork it, sell it, self-host it
• No accounts, no tracking, no cloud backend

Caveats

• The team is now primarily building char; anarlog is maintained but not the main focus
• Self-hosting requires building from source—no one-click deploy docs are visible
• It’s unclear how polished the local transcription quality is versus cloud alternatives

Verdict

Worth a look if you’re privacy-paranoid, want your meeting history in git, or need a Granola-shaped thing without the Granola pricing. Skip it if you want a polished SaaS with support SLAs and zero setup friction.