This module is designed for asynchronous usage and provides a simple interface to the Google Gemini API.

It also provides a command-line interface for interactive usage.

Images are not tested and may not work as expected.

pip install -U gemnine

gemnine requires a model and an API key to be set.

You can use config file, command parameters or interactive input to set these values.

By default, gemnine will look for files named config.json and session.json in the user config and cache directory.

gemnine -c /path/to/config.json  # Use a config file
gemnine -m "models/gemini-pro"  # Set model, input API key interactively
gemnine -l  # list available models

• -c, --config: Path to a JSON file
• -m, --model: Model name
• -s, --session: Session history file
• -V, --version: Show version

Precedence: Interactive input > Command parameters > Config file

This mode mimics a chat interface. You can type your message and get a response.

Commands are started with a / and are case-sensitive.

• /exit: Exit the program
• /save <path>: Save the session history to a file
• /load: Load a session history from a file
• /rollback <step>: Rollback to the previous state
• /clear: Clear the session history
• /role: Switch role
• /model: Switch model
• /help: Show help

To embed an image, insert #image(<url>) in your message. Use double # to escape embedding.

await bot.send("""
What are these?
#image(https://example.com/image.png)
#image(file:///path/to/image.png)
#image(base64://<base64-encoded-image>)
""")

All URLs will be downloaded and embedded as base64-encoded images.

import asyncio
import gemnine

bot = gemnine.Bot(model="models/gemini-pro", api_key="your-api-key")

async def main():
    response = await bot.send("Hello, how are you?")
    print(response)

    async for r in bot.stream("I'm fine, thank you."):
        print(r, end='')
    
    # `Bot` instance it self doesn't track the session history
    # To keep track of the session history, use `Session` instance
    sess = bot.new_session()

    print(await sess.send("Hello, how are you?"))

    async for r in sess.stream("What was my last question?"):
        print(r, end='')

asyncio.run(main())

Session and Bot use pydantic models under the hood. You can save and load the session history using model_dump_json and model_validate_json or pass arguments to new_session methods.

import json
from pathlib import Path
sess_path = Path("session.json")

sess = bot.new_session()

data = sess.model_dump_json()

sess_path.write_text(data)

sess = bot.new_session(**json.loads(sess_path.read_text()))