TALAX
GitHub Back to overview
Build from source

Install TalaX and validate the local dictation workflow.

There are no published binary releases yet. The documented install path today is a source build using Rust, Node.js, Tauri, and the current engine/frontend workspace.

1. Prerequisites

Start with the documented development stack from the repository. Only claim support you can verify; the current published setup explicitly documents Linux system packages.

  • Rust (stable, edition 2024)
  • Node.js 18+
  • Linux system packages: libasound2-dev, libx11-dev, libxtst-dev, cmake, pkg-config

2. Clone the repository

git clone https://github.com/puretensor/TalaX.git
cd TalaX

If you are evaluating the product rather than contributing, this still gives you the clearest view of the current state until packaged downloads exist.

3. Install frontend dependencies and run the app

cd ui
npm install
cd ..
cargo tauri dev

This runs the Tauri desktop app locally so you can test hotkey capture, transcription, correction, and text injection end to end.

4. Validate the engine and frontend

cargo test -p pureclaw-engine
cd ui
npx svelte-check

Run the engine tests before trusting changes and use the frontend type checks to catch UI drift early.

5. Choose a Whisper model

The model manager downloads Whisper models on first use from HuggingFace with progress tracking and integrity verification.

  • tiny.en — ~75 MB — fastest
  • base.en — ~142 MB — fast
  • small.en-q5_1 — ~181 MB — recommended balance
  • medium.en-q5_0 — ~515 MB — higher accuracy
  • large-v3-turbo-q5_0 — ~574 MB — highest accuracy / slowest

6. What to validate on first run

  • The global hotkey triggers reliably from your target applications.
  • Whisper downloads and transcribes without sending audio to a remote service.
  • Correction output lands correctly in your editor or chat client.
  • Repeated fixes for your domain terms start improving the dictated output.