What "self-hosting Earshot" actually means

Earshot's desktop app is the same whether you self-host or subscribe. Self-hosting just means you supply the AI instead of us: you connect your own accounts for the two AI steps — turning speech into text (transcription) and turning text into notes (the language model) — and you pay those providers directly. The app, the recording, and your data all stay on your Windows PC. It's free forever, and the code is on GitHub under the MIT licence if you want to read or change it.

What you'll need

A Windows 10 or 11 PC (64-bit).
A transcription option — see the choices below (there's a genuinely free path).
An AI model for the notes — a paid API key, or a local model for zero ongoing cost.

Step 1 — Download and install

Grab the installer from the download page (or straight from GitHub Releases) and run it. The installer isn't code-signed yet, so Windows SmartScreen may ask you to confirm — choose More info → Run anyway. Because Earshot is open source, you can check exactly what you're installing, or build it yourself from the repo.

Step 2 — Pick a transcription provider

This is the speech-to-text step. Your options, from most private to most convenient:

Your own Whisper server (free, fully local). Run Whisper on your machine and the audio never leaves it. The most private option, and free.
Groq. A generous free tier, then around $0.04 per hour of audio — fast and cheap.
Deepgram or OpenAI. Solid managed transcription if you already have an account.

Step 3 — Pick an AI provider for the notes

This is the step that writes the summary, decisions, action items and answers.

Anthropic (Claude) — paste an API key.
Any OpenAI-compatible endpoint — use OpenAI, or another provider that speaks the same API.
A fully local model via Ollama or LM Studio — on this setup, your meeting text never leaves your machine at all.

Step 4 — Paste your keys

On first run, Earshot's setup guide walks you through connecting the providers you chose. Keys are stored locally. Prefer documentation? The GitHub README covers every option in detail, including pointing Earshot at your own Whisper server.

Step 5 — Record

Hit record — or let call detection offer when it notices a meeting start — and Earshot captures both sides of the conversation locally, even for hours-long sessions. When it's done you get a tight summary, the decisions, and suggested action items you approve or dismiss. Everything lands in a folder on your PC.

Self-host or Plus? A quick honest take

Self-hosting is the right call if you want maximum control, a fully local pipeline, or simply free. The trade-off is that you manage keys and providers yourself. If you'd rather not — no keys, no configuration, managed transcription and AI, plus cloud features as they roll out — Earshot Plus is US$9 a month with a 7-day free trial, and you can switch between the two in either direction anytime without losing a thing. Either way, the recording still happens locally with no bot in your call. For the privacy reasoning behind all this, see local vs cloud transcription.

Self-hosting Earshot