Gemma Firecrawl Agent

Small Python ADK-style scaffold for an AG2 agent that talks to a local Ollama model and can call Firecrawl to scrape web pages into LLM-ready text.

Setup

After this project is published to PyPI:

pip install gemma-firecrawl-agent

For local development from source:

python -m venv .venv
source .venv/bin/activate
pip install -e .
cp .env.example .env

Edit .env:

FIRECRAWL_API_KEY=fc-your-key
OLLAMA_MODEL=gemma3:4b
OLLAMA_VLM_MODEL=qwen3-vl:8b
OLLAMA_HOST=http://127.0.0.1:11434
LLAMA_CPP_MODEL=llama-cpp
LLAMA_CPP_BASE_URL=http://127.0.0.1:8080/v1
LLAMA_CPP_API_KEY=sk-no-key-required

Make sure Ollama is running and the model is pulled:

ollama pull gemma3:4b
ollama pull qwen3-vl:8b
ollama serve

Run

gemma-firecrawl-agent "Scrape https://example.com and summarize the page."

Use the VLM provider profile:

gemma-firecrawl-agent --provider vlm "Scrape https://example.com and summarize the page."

Use a local llama.cpp server:

llama-server -m /path/to/model.gguf --host 127.0.0.1 --port 8080
gemma-firecrawl-agent --provider llama-cpp "Scrape https://example.com and summarize the page."

Use structured output when you want a JSON object:

gemma-firecrawl-agent --structured "Scrape https://example.com and summarize the page."

Structured output also works with the VLM provider:

gemma-firecrawl-agent --provider vlm --structured "Scrape https://example.com and summarize the page."

And with llama.cpp:

gemma-firecrawl-agent --provider llama-cpp --structured "Scrape https://example.com and summarize the page."

Sample structured output shape:

{
  "title": "Short page or task title",
  "url": "https://example.com",
  "summary": "One concise paragraph.",
  "key_points": [
    "First important point",
    "Second important point"
  ],
  "source_status": "scraped"
}

Print the sample shape from the CLI:

gemma-firecrawl-agent --show-structured-shape

You can also run the module directly:

python -m gemma_firecrawl_agent.cli "Scrape https://example.com and list the key facts."

Publishing

Release steps live in PUBLISHING.md. The short version is:

pip install -e ".[dev]"
python -m build
python -m twine check dist/*