Gemini Guide¶

First-time setup¶

Run setup once before using Gemini. This opens a browser so you can build a real user profile and optionally log in:

from hermex import Gemini

Gemini.setup()

Browse around briefly, log in to your Google account if you want image upload support, then close the window.

Guest mode vs logged-in mode¶

Gemini supports two modes:

Feature	Without login	Logged in
Text queries	✓	✓
File upload	✗	✓
Image generation	✓	✓

If you try to upload files without being logged in, Hermex raises a LoginRequiredError.

Basic text query¶

from hermex import Gemini

gemini = Gemini()
gemini.open_url()

response = gemini.query("Explain how transformers work in simple terms.")
print(response.text)

gemini.close()

Attaching files¶

File upload requires a logged-in session. Supported formats: .jpg, .jpeg, .png, .gif, .webp, .pdf, .csv, .txt, .json. You can also check Gemini.SUPPORTED_ATTACHMENTS at runtime.

response = gemini.query(
    "What's wrong with this code?",
    attachments=["screenshot.png"],
)
print(response.text)

You can attach multiple files:

response = gemini.query(
    "Compare these two diagrams.",
    attachments=["diagram1.png", "diagram2.png"],
)

Getting generated images¶

When Gemini generates an image, Hermex downloads it automatically and returns the path:

response = gemini.query("Generate an image of a futuristic city at night.")
if response.image:
    print(f"Image saved to: {response.image}")

By default images are saved to the current working directory. Change this with download_dir:

gemini = Gemini(download_dir="outputs/")

Watermark removal¶

Gemini watermarks its generated images. Pass remove_watermark=True to strip it automatically:

response = gemini.query(
    "Generate an image of a sunset over the ocean.",
    remove_watermark=True,
)

This uses OpenCV template matching against the known watermark asset — no external service involved.

Getting markdown¶

By default query() returns plain text. Pass get_markdown=True to get the raw markdown source instead:

response = gemini.query(
    "Write a comparison table of Python web frameworks.",
    get_markdown=True,
)
print(response.text)  # raw markdown with table syntax

Long messages¶

For long prompts, character-by-character typing is slow. Use paste=True to paste the message instead:

response = gemini.query(long_prompt, paste=True)

By default, Hermex types some dummy text first before pasting (fake_typing=True) to avoid bot detection. You can disable this if needed:

response = gemini.query(long_prompt, paste=True, fake_typing=False)

Multi-turn conversation¶

query() always appends to the existing conversation. Just call it multiple times:

gemini.open_url()

gemini.query("You are a helpful assistant. Let's work through a problem step by step.")
response = gemini.query("What is the time complexity of quicksort?")
print(response.text)

response = gemini.query("Now explain why the worst case happens.")
print(response.text)

One-shot scripts¶

For scripts that just need a single answer, simple_query() handles everything:

from hermex import Gemini

response = Gemini.simple_query("Translate 'hello world' to Spanish.")
print(response.text)

Constructor options¶

gemini = Gemini(
    download_dir="outputs/",   # where to save generated images
    headless=False,            # set True to run without a visible window
    typing_delay=0.025,        # seconds between keystrokes (default 0.025)
    data_dir=None,             # override the default data directory
)

Warning

Avoid headless=True for sessions where bot detection is a concern. A visible browser is significantly harder to detect.