🔥 spark

A CLI for interacting with the DGX Spark llama.cpp server.

Quick access to your local LLM from the terminal — ask questions, chat, generate tweets, summarize files, and more.

Installation

# Clone and install
git clone https://git.cataco.net/Catacolabs/spark-cli.git
cd spark-cli
./install.sh

# Or just copy the script
cp spark ~/.local/bin/
chmod +x ~/.local/bin/spark

Usage

# Check if Spark server is running
spark status

# Ask a question
spark ask "What are the key features of Rust?"

# Interactive chat
spark chat

# Generate tweet ideas
spark tweet "Workers AI announcements"

# Summarize a file
spark summarize paper.pdf
cat notes.txt | spark summarize

# Generate code
spark code "async HTTP server in Python with FastAPI"

# Text completion (non-chat)
spark complete "The future of AI is"

Commands

Command	Description
status	Show server health and loaded model
ask <question>	Ask a single question, get a response
chat	Interactive chat mode with history
tweet <topic>	Generate 3 tweet ideas about a topic
summarize <file>	Summarize text from file or stdin
code <description>	Generate code from a description
complete <text>	Raw text completion (non-chat)
raw <endpoint>	Make raw API calls (debugging)

Environment Variables

Variable	Default	Description
SPARK_URL	http://spark:8080	llama.cpp server URL
SPARK_MODEL	(auto-detected)	Model to use for completions

Requirements

• curl - HTTP client
• jq - JSON processor
• A running llama.cpp server on your DGX Spark

Examples

Quick Research

# Get a quick explanation
spark ask "Explain transformer attention mechanisms in simple terms"

# Summarize a paper
curl -s "https://arxiv.org/abs/2301.00000" | spark summarize

Content Creation

# Tweet ideas for your content pipeline
spark tweet "new ComfyUI workflow for video generation"

# Expand on an idea
spark ask "Write a thread about why local LLMs matter for privacy"

Coding Help

# Quick code generation
spark code "Python function to parse RSS feeds and extract titles"

# Debug help
echo "TypeError: 'NoneType' object is not subscriptable" | spark ask "What causes this error?"

Interactive Session

$ spark chat
🔥 Spark Chat (model: Qwen2.5-72B-Instruct)
Type 'exit' or Ctrl+C to quit, '/clear' to reset

You> What's the difference between async and threading in Python?
Spark> [response...]

You> Give me an example of when to use each
Spark> [response...]

You> /clear
Chat history cleared

You> exit
Goodbye!

How It Works

The CLI communicates with llama.cpp's OpenAI-compatible API:

• /v1/chat/completions for chat-based commands
• /v1/completions for raw text completion
• /v1/models to detect the loaded model
• /health for status checks

Tips

1. Server offline? Make sure your DGX Spark is powered on and llama.cpp is running
2. Slow responses? Large models take time — the CLI has a 120s timeout
3. Want streaming? Not yet implemented, but PRs welcome!
4. Custom model? Set SPARK_MODEL=your-model-name

---

Built during an overnight session by Navi ✨