Requirements
Hardware
| Requirement | |
|---|---|
| Computer | A Mac with Apple Silicon (M1 or later). setup.sh refuses to run on other machines. |
| macOS | macOS 14 Sonoma or later |
| Memory | 16 GB recommended. Speech models, the browser and an Ollama model all share memory while they run |
| Disk | About 5 GB for the build, Node packages and speech models, plus your recordings (see below) |
Software
| Tool | Why |
|---|---|
| Xcode command line tools | Compilers for the Rust server |
| Rust (via rustup) | Builds the server and the web app. setup.sh adds the WebAssembly target and the pinned wasm-bindgen-cli |
| Node.js 22.18 or newer | Builds the web assets, and runs Whisper Large V3 at runtime |
| pnpm | Installs the JavaScript dependencies |
| Brave or Chrome 120+ | Runs the Echo web app and the recording extension |
| Ollama (optional) | Generates meeting notes locally |
Install walks through each one on a fresh Mac.
Disk space for recordings
Echo stores meeting audio as encrypted 16 kHz mono PCM, about 115 MB per hour of recording. Uploaded files are stored as they are, so their size depends on the format. Transcripts and notes are small by comparison.
| Item | Size |
|---|---|
| One hour of recorded meeting audio | about 115 MB |
| Whisper Large V3 Turbo, with the speaker models | about 1.2 GB, cached in the browser |
| Whisper Large V3 | about 1.7 GB, under the data folder |
A recommended Ollama model (qwen2.5:3b, llama3.2:3b, gemma3:4b) | about 2–3.5 GB each, stored by Ollama |
Models are downloaded once, when you choose to. Echo never downloads them on its own.
Network
Echo needs the internet only to build it and to download the models you pick. After that, recording, transcription and notes work offline. See privacy for every request Echo can make.