CyberCat

CyberCat is a small Nx workspace with:

• a Fastify service that discovers runnable script commands and streams output over Server-Sent Events
• a React web app that lets you filter, run, and watch those commands in a terminal-style UI
• a global shortcut selection speaker that captures selected text and asks an OpenAI-compatible qwen-omni model to speak it
• a global shortcut selection assistant that can capture selected text, call an OpenAI-compatible chat endpoint, stream the result into the web UI, and append an output-only Markdown log
• colocated command entrypoints under commands and a local .env file in the CyberCat root

Structure

• apps/service contains the Fastify API and command execution backend
• apps/web contains the React UI and web assets
• commands/xgd contains the xgd test scripts
• commands/zixiCat contains the zixiCat helper scripts

Development

Create a local environment file from .env.example and add your credentials before running the xgd commands.

Start the web app and service together:

npm start

Start the service:

npx nx serve @cyber-cat/service

Start the web app:

npx nx serve @cyber-cat/web

Build both apps:

npx nx build @cyber-cat/service
npx nx build @cyber-cat/web

Selection Speaker

The selection speaker listens for a global shortcut, reads the current Windows selection, requests audio from an OpenAI-compatible chat completions endpoint, writes a temporary WAV file, and plays it locally.

Configure it in .env.example through these variables:

• SELECTION_SPEAKER_SHORTCUT to change the global shortcut
• SELECTION_SPEAKER_API_KEY, SELECTION_SPEAKER_BASE_URL, and SELECTION_SPEAKER_MODEL for the audio-capable LLM endpoint
• SELECTION_SPEAKER_VOICE to choose among the currently supported qwen-omni voices: Ethan, Chelsie, or Aiden
• SELECTION_SPEAKER_MAX_INPUT_LENGTH to cap the captured selection before sending it upstream

Selection Assistant

The selection assistant starts with the service on Windows, listens for a global shortcut, reads the current selection, sends it through the OpenAI JavaScript SDK to an OpenAI-compatible /chat/completions endpoint, shows the latest result in the web UI, and appends each output to a local Markdown log.

Configure it in .env.example through these variables:

• SELECTION_ASSISTANT_SHORTCUT to change the global shortcut
• SELECTION_ASSISTANT_API_KEY, SELECTION_ASSISTANT_BASE_URL, and SELECTION_ASSISTANT_MODEL for the LLM endpoint
• SELECTION_ASSISTANT_PROMPT_PATH to override the default prompt file at apps/service/src/assets/selection-assistant.prompt.md
• SELECTION_ASSISTANT_LOG_PATH to choose the output-only Markdown log destination

Commands

CyberCat discovers shell scripts directly from commands/xgd and commands/zixiCat. They run from the repository root:

bash ./commands/xgd/qwen-chat-llm-test.bash
bash ./commands/xgd/qwen-embed-test.bash
bash ./commands/xgd/qwen-rerank-test.bash
bash ./commands/zixiCat/copy-command2run-remote-bashrc.bash