Ways to use THETA¶

English | 中文

Choose the desktop app for installation without a development environment, the Web workbench for a local browser, or the CLI Agent for terminal conversations. All three use local project storage and the same analysis engine.

Desktop app¶

Download the installer from GitHub Releases. Use the DMG for Apple Silicon Macs and the EXE for Windows x64. On macOS, drag THETA into Applications; on Windows, run the installer.

Python and CPU compute dependencies are included. You do not need to install Python, Node.js, or Conda. Model weights and user credentials are not bundled. This is an unsigned preview; your operating system may show a security prompt.

Open Settings inside THETA to enter your conversation model endpoint, model name, and API key. Local embeddings accept a compatible model directory, with Qwen3-Embedding-0.6B as the default suggestion. Cloud embeddings accept your endpoint and key, with GLM embedding-3 as the preset. THETA zero-shot, CTM and BERTopic support cloud embeddings. THETA fine-tuning requires compatible local weights.

Launch does not require a Keychain password. API keys use local user-owned encryption; users upgrading from legacy Keychain storage re-enter their keys once.

Projects and results remain on your computer. macOS uses ~/Library/Application Support/THETA. New Windows installations use a writable THETA-data folder inside the installation directory, falling back to %APPDATA%/THETA when needed. Updates preserve the existing data location.

Skill Manager¶

In conversation mode, open Skill Manager above New project. Import a folder, ZIP, tar.gz or SKILL.md, or download a skill from a GitHub repository or skill folder URL. Enable or disable skills, export packages and remove imported skills. Built-in Data Viz plotting templates and Agent-requested downloads appear in the same list. The Agent reads enabled skill instructions before use; importing files does not execute scripts or install dependencies.

Web workbench¶

For source use, install Node.js 22.13+, pnpm, and the Python compute dependencies. Create private agent/.env.local from the example only on first setup, then configure your model endpoint and key. Preserve existing configuration.

From the repository root:

pnpm --dir agent install --frozen-lockfile
npm --prefix frontend ci
./theta-web start

Open http://127.0.0.1:4320/workbench. Choose conversation or manual mode at the top. Use ./theta-web status to check services and ./theta-web stop to stop them.

The launcher starts local services and preserves projects in .theta_agent/ and .local/manual-workbench/. Keep these directories when updating source code.

CLI Agent¶

After source setup, run ./theta from the repository root. On Windows, use node agent/cli/bin/theta.mjs. Describe your question and attach a data file. Review model settings and confirm the proposed computation before execution.

Use /help for commands, /sessions to find conversations, and /resume to continue one. See the CLI guide for configuration and command details.

Workflow skill¶

Compatible agents can read skills/theta-workflow/SKILL.md to guide data preparation, model selection, execution confirmation, and result interpretation. See the workflow instructions.