Unmaintained. Use google-genai, Google's unified Gen AI SDK for Python, or Jules if you want a terminal-first coding agent.

I've always preferred pair programming with LLMs from the terminal over copy-pasting from browser chats. Web UIs hit rate limits, lose context across tabs, and break flow when I'm moving code back and forth. Google shipped the Gemini API in mid-December 2023 and the pitch was native multimodal: one request could handle text, images, video, audio, PDFs, and code together. I shipped this two weeks later. The first-party SDK was fresh, and I wanted all three modes (chat / text / multimodal), URL-or-local file uploads, streaming, JSON output, and the full parameter + safety surface behind one CLI. So I built this for myself.

Python wrapper and CLI for Google's Gemini models, with native multimodal.

git clone https://github.com/ramonclaudio/gemini-ai-toolkit.git
cd gemini-ai-toolkit
pip install -r requirements.txt

Get an API key at https://aistudio.google.com/app/apikey. Set via env var, .env, or pass directly:

export GEMINI_API_KEY=your_api_key

# Chat
python cli.py --chat

# Text
python cli.py --text --prompt "Write a story about a magic backpack."

# Multimodal (files from local paths and/or URLs)
python cli.py --multimodal --prompt "Summarize each file." --files file1.jpg https://example.com/file2.pdf

Type exit or quit to leave chat.

from gemini import Chat, Text, Multimodal

Chat().run()
Text().run(prompt="Write a story about a magic backpack.")
Multimodal().run(prompt="Summarize each file.", files=["file1.jpg", "https://example.com/file2.pdf"])

• /exit or /quit: end the session
• /clear: clear conversation history (saves API credits)
• /upload: (multimodal) /upload file_path_and_or_url [optional prompt]

Google's Files API stores up to 20 GB per project, 2 GB per file, 48-hour retention.

URL-downloaded files cache to .gemini_ai_toolkit_cache and clear on session end.

Rate limits (429) auto-retry after 15s.

MIT