Hello world!

Welcome

Bristlenose is in early alpha. It's being built in the open by one person, with a small group of friends kicking the tyres. Things will be rough around the edges, and we want to hear about it when they are.

How to try

On a Mac — TestFlight alpha (easiest)

The desktop app is in TestFlight alpha right now. It bundles Python, FFmpeg, and Whisper — no terminal needed. You bring an API key for one of the cloud AI providers, or run a local model via Ollama, and that's it.

Ping me for an invite and I'll add you. Requires Apple Silicon (M1 or later) and macOS 15 Sequoia.

On Linux, or Mac via the command line

The CLI version of Bristlenose is open source and stable, and runs the same pipeline the desktop app does. Pick whichever fits your setup:

# macOS via Homebrew — handles Python and FFmpeg for you
brew install cassiocassio/bristlenose/bristlenose

# Any platform via pipx
pipx install bristlenose

# Any platform via uv
uv tool install bristlenose

If you go the pipx or uv route you'll also need FFmpeg installed separately: brew install ffmpeg on macOS, or sudo apt install ffmpeg on Ubuntu/Debian.

Then run bristlenose doctor to confirm everything is wired up. The manual has the full walkthrough — installation, AI-provider setup, your first run, and the CLI commands.

A note on coverage: macOS is the primary platform. The CLI works on Linux too, though it's less heavily exercised — if you run it there and hit something, that's especially valuable feedback.

Get help

The fastest path is a GitHub issue — that's where bugs, questions, and feature requests all land. No issue is too small.

If you're using the Mac app, there's a feedback button in the app itself — that comes through to the same place.

Or just email hello@bristlenose.app.

Feedback

Anything at all is useful, but if you're looking for nudges:

  • The AI providers. We want to exercise all the public LLMs — Claude, ChatGPT, Azure OpenAI, Gemini — and the local-model path via Ollama. If one feels noticeably better or worse than another for your data, that's gold.
  • Recording formats. Different versions of Zoom and Teams produce different file formats, and some of them break things in ways we haven't seen yet. Same goes for other audio and video formats — anything you have lying around is fair game.
  • Length and size. Does it cope with very long interviews? Large screen recordings? Whole-day workshops? We don't have enough variety in our test data to know.
  • The default codebooks. Bristlenose ships with a handful of UX-research codebooks (frameworks for tagging quotes). Which ones are useful to you, which ones get in the way, what's missing?
  • Themes and signals. Do you trust the themes it surfaces? The sentiment classification? The signal cards on the analysis page? Where does it feel right, where does it feel off?
  • UX rough edges. Anywhere it's slow, confusing, or makes you reach for the manual. Especially the first-run flow — that's the part we're least sure about.

Thanks.

— Martin