Humla

Humla is a Mac app that records meetings without sending a bot into the call. The other side sees nothing. Audio is captured through macOS itself, transcribed live, and summarized the moment you're done. No sign-up, no cleanup, no cloud account required.

It's built for freelancers, consultants, and small teams who take client calls and need a written record fast. The core workflow is three steps: jot an agenda before, record during, copy the summary after. A five-minute meeting takes the same effort as an hour-long one.

Key capabilities:

• Multiple summary presets, the same transcript becomes a client recap, an internal action list, or a custom format. Switch presets and regenerate without re-recording.
• Chat with your notes, a chat pane searches your full meeting history by meaning, not just keywords, and cites the exact note it pulled from. Works fully offline via Ollama, or with your own OpenAI key.
• On-device transcription, local Whisper handles around 100 languages with auto-detection. Speaker ID also runs locally, free, without uploading audio.
• Bring your own model, route transcription to Whisper, Deepgram, or Groq. Use local Ollama models for summaries or plug in your own API key.
• Custom vocabulary, teach it names, acronyms, and jargon specific to your work.
• Click-to-edit transcript, fix a word directly; playback stays in sync.

Compared to tools like Granola, Otter.ai, or Fireflies, Humla's privacy posture is different by design. Those are cloud products: your recordings live on their servers. Here, everything stays on your Mac unless you explicitly choose a cloud provider on your own account. API keys are stored in the macOS Keychain, never in plaintext. The app collects no user ID and sends no meeting data anywhere.

Team sync is available through Humla Cloud at $5 per user per month, or you can self-host it against your own PocketBase server for free. Sync handles notes and workspaces only; transcription and summaries still run on your own providers or on-device either way.

The app is MIT-licensed and fully auditable. It runs on Apple Silicon Macs (macOS 13+) and has no Windows or web version yet.