TARS DESKTOP 2.0 - WINDOWS SETUP

This ZIP is the full local AI edition. It is source-based, not a one-click
installed executable. Node.js, Ollama and model weights are NOT included.
Nothing in this package installs dependencies, downloads a model, enables
cloud AI, or configures automatic startup by itself.

ONE-TIME SETUP
1. Right-click the ZIP and choose Extract All. Keep the extracted folder in
   a writable location, such as Documents. Do not launch from inside the ZIP.
2. Install Node.js 24 LTS for Windows: https://nodejs.org/en/download
   Node.js 22 or later is required. Leave the installer PATH option enabled.
3. Install Ollama for Windows: https://ollama.com/download/windows
   Windows 10 22H2 or newer is required by Ollama. Start Ollama normally.
4. Open a new PowerShell window and run:

       ollama pull qwen2.5:3b

   This is a separate model download of approximately 1.9 GB. Allow storage
   for Ollama as well as its models. Model terms are supplied by its publisher:
   https://ollama.com/library/qwen2.5:3b
   TARS can also use other chat models you already installed in Ollama.
5. Double-click Launch TARS.vbs in the extracted TARS-Desktop-v2 folder.
   If your Windows configuration disables VBScript, use Launch TARS.cmd.
   The app opens in your default browser. Wait for LOCAL LINK ACTIVE.

DAILY USE
- Keep Ollama running. Open Launch TARS.vbs (or Launch TARS.cmd).
- Ask a question. Humor, Brevity and Formality change the next real response.
- Stop TARS.vbs (or Stop TARS.cmd) closes this app's server and leaves Ollama
  and other programs running. Closing the browser alone leaves TARS running.
- No administrator session, npm installation, API key, or cloud account is
  needed by TARS. Node/Ollama installers have their own setup requirements.

IF LAUNCH DOES NOT WORK
- In a fresh PowerShell window, check: node --version
- Check installed models with: ollama list
- Start Ollama, then use Reconnect in TARS preferences.
- Look in .runtime/launcher-error.log or .runtime/server-error.log after a
  failed launch. These files are created locally; none are shipped here.
- As a fallback, open PowerShell in the extracted folder and run:

       node server.mjs

  Then open http://127.0.0.1:43710/ and use Ctrl+C to close that manual server.
  If its port is occupied, use the provided launcher to choose a free port.

PRIVACY AND LIMITS
TARS listens only on this computer's loopback address. Chat requests go to
local Ollama. It has no tools, live web access, file access or computer control.
The package does not enable or call a cloud inference service. Choose a local
Ollama model. First-time software/model downloads need an internet connection.

Chats/settings are saved in your browser for this exact hostname and port.
Use Export before clearing browser data, moving to a different browser or
changing ports; use Import to restore an export. No personal chats are bundled.

Optional speech uses installed Windows voices and Windows speech recognition.
It is not the film actor's voice. Microphone permission, installed language
packs and recognition quality vary. Review transcripts before sending.
Text chat remains usable when speech is unavailable. The 3D scene requires a
WebGL-capable browser; reduced motion and lower effects are available.

See README.md for the controls, data limits and optional developer checks.
This is an independent, noncommercial Interstellar-inspired fan project.
