BuzzLLM Gateway

What is BuzzLLM

A CLI gateway that invokes LLMs with specialized tools:

• websearch: Search the web and scrape pages
• pythonexec: Execute Python code in a sandboxed Docker container
• codesearch: Search and read files in local codebases
• hackhub: Generate Search-Replace blocks for code modifications

Command Structure

buzzllm "MODEL" "API_URL" "PROMPT" \
    --provider PROVIDER \
    --api-key-name ENV_VAR \
    --system-prompt TEMPLATE \
    [--think]  # for extended thinking
    [--brief]  # hide tool calls/results, show only final output

Providers

System Prompt Templates

With Tools

• websearch - Web search + page scraping (search_web, scrape_webpage)
• codesearch - Codebase analysis (bash_find, bash_ripgrep, bash_read)
• pythonexec - Python in Docker (python_execute)

Without Tools

• hackhub - Search-Replace block generation
• generate - General generation
• helpful - Helpful assistant
• replace - Replacement template

Usage Examples

Web Search

buzzllm "gpt-4o-mini" \
    "https://api.openai.com/v1/chat/completions" \
    "What is the current price of Bitcoin?" \
    --provider openai-chat \
    --api-key-name OPENAI_API_KEY \
    --system-prompt websearch

Python Execution

# Requires Docker container: cd python_runtime_docker && bash build_docker.sh build-python-exec
buzzllm "claude-sonnet-4-20250514" \
    "https://api.anthropic.com/v1/messages" \
    "Calculate the first 20 Fibonacci numbers using Python" \
    --provider anthropic \
    --api-key-name ANTHROPIC_API_KEY \
    --system-prompt pythonexec

Code Search

buzzllm "gpt-4o-mini" \
    "https://api.openai.com/v1/chat/completions" \
    "How does error handling work in this codebase?" \
    --provider openai-chat \
    --api-key-name OPENAI_API_KEY \
    --system-prompt codesearch

Code Modifications (HackHub)

buzzllm "claude-sonnet-4-20250514" \
    "https://api.anthropic.com/v1/messages" \
    "$(cat src/main.py)\nAdd logging to all functions" \
    --provider anthropic \
    --api-key-name ANTHROPIC_API_KEY \
    --system-prompt hackhub

Extended Thinking (Claude/OpenAI reasoning)

buzzllm "claude-sonnet-4-20250514" \
    "https://api.anthropic.com/v1/messages" \
    "Solve this complex problem step by step" \
    --provider anthropic \
    --api-key-name ANTHROPIC_API_KEY \
    --think

When to Use

1. User asks for web search: Use --system-prompt websearch
2. User wants Python execution: Use --system-prompt pythonexec (ensure Docker is running)
3. User asks about codebase: Use --system-prompt codesearch
4. User wants code changes: Use --system-prompt hackhub
5. Complex reasoning needed: Add --think flag

Environment Variables Required

• OPENAI_API_KEY - For OpenAI models
• ANTHROPIC_API_KEY - For Anthropic models
• BRAVE_SEARCH_AI_API_KEY - Fallback for websearch if DuckDuckGo fails

Output

Default: streams text to stdout With -S flag: SSE format for programmatic consumption