---

✨ Features

• Free access — Uses the Gemini web UI, no API key or billing required
• Model selection — Switch between Fast, Thinking, and Pro models
• Async & Sync APIs — Use GeminiClient (async) or GeminiClientSync (sync)
• Easy auth — Log in via your regular browser cookies, no manual sign-in needed
• Session persistence — Log in once, reuse the session across runs
• Headless mode — Run without a display after initial setup
• Anti-detection — System Chrome + stealth measures to bypass automation detection
• Multi-turn conversations — Continue chats or start new ones
• Response streaming detection — Waits for the full response automatically

📦 Installation

# Clone the repo
git clone <your-repo-url>
cd gemini-api-again

# Install the package
pip install -e .

# Install Playwright browsers (first time only)
playwright install chromium

# Optional: Install cookie extraction support
pip install -e ".[cookies]"

Note: You also need Google Chrome installed on your system. The library uses your system Chrome to avoid Google's "browser not secure" detection.

# Ubuntu/Debian
sudo apt install google-chrome-stable

🚀 Quick Start

Option A — Login from your browser (easiest, no manual sign-in)

Extract cookies from your regular browser — no Playwright sign-in window needed:

# Install cookie extraction support
pip install -e ".[cookies]"

import asyncio
from gemini_webui import GeminiClient

async def main():
    async with GeminiClient(headless=True, session_path="sessions/default.json") as client:
        # Extract cookies from Chrome — you must be logged into
        # gemini.google.com in your Chrome browser
        await client.login_from_browser(browser_name="chrome")

        response = await client.send_prompt("Explain quantum computing in 3 sentences")
        print(response.text)

asyncio.run(main())

Supported browsers: "chrome", "chromium", "firefox", "opera", "edge"

Option B — Manual login via Playwright browser window

Run this once to log in and save your session. A Chrome window opens — sign in with your Google account:

import asyncio
from gemini_webui import GeminiClient

async def setup():
    async with GeminiClient(headless=False, session_path="sessions/default.json") as client:
        await client.login()  # Browser opens — log in manually
        # Session saved automatically on exit

asyncio.run(setup())

After that, use headless mode with the saved session:

async with GeminiClient(headless=True, session_path="sessions/default.json") as client:
    await client.login()
    response = await client.send_prompt("Explain quantum computing in 3 sentences")
    print(response.text)

Output:

Quantum computing is a revolutionary technology that uses the principles of quantum
mechanics to process information in ways traditional computers cannot. Instead of
standard bits that represent either a 0 or a 1, quantum computers use qubits, which
can exist as both 0 and 1 simultaneously thanks to a property called superposition.
This unique ability allows them to perform complex calculations at unprecedented speeds.

Sync API (simpler, no async/await)

from gemini_webui import GeminiClientSync

with GeminiClientSync(headless=True, session_path="sessions/default.json") as client:
    # Option A: From browser cookies
    client.login_from_browser(browser_name="chrome")

    # Option B: From saved session
    # client.login()

    response = client.send_prompt("What is the meaning of life?")
    print(response.text)

🔄 Model Selection

Switch between Gemini models — Fast (Flash), Thinking, and Pro:

from gemini_webui import GeminiClientSync

with GeminiClientSync(headless=True, session_path="sessions/default.json") as client:
    client.login()

    # List available models
    models = client.list_models()
    print(models)  # ['Fast', 'Thinking', 'Pro']

    # Check current model
    print(client.get_current_model())  # 'Fast'

    # Switch to Pro for advanced tasks
    client.set_model("pro")
    response = client.send_prompt("Solve: integral of x^2 * e^x dx")
    print(response.text)

    # Switch to Thinking for complex reasoning
    client.set_model("thinking")
    response = client.send_prompt("What is 15! / (3! * 5! * 7!)?")
    print(response.text)

    # Switch back to Fast for quick answers
    client.set_model("fast")

Model	Alias	Description
Fast	"fast", "flash"	Quick answers (Gemini Flash)
Thinking	"thinking"	Solves complex problems with reasoning
Pro	"pro"	Advanced math and code (Gemini Pro)

💬 Multi-Turn Conversations

from gemini_webui import GeminiClientSync

with GeminiClientSync(headless=True, session_path="sessions/default.json") as client:
    client.login()
    client.new_chat()  # Start a fresh conversation

    # First message
    r1 = client.send_prompt("Tell me about Python in 2 sentences")
    print(r1.text)

    # Follow-up — continues in the same conversation
    r2 = client.send_prompt("What are its main advantages?")
    print(r2.text)

    # Start a completely new conversation
    r3 = client.send_prompt("What is Rust?", new_chat=True)
    print(r3.text)

🎛️ Configuration

client = GeminiClient(
    headless=True,              # Run browser in headless mode
    timeout=120.0,              # Response wait timeout (seconds)
    session_path="sessions/default.json",  # Session persistence path
    browser_type="chromium",    # Browser engine (ignored if use_system_chrome=True)
    slow_mo=0,                  # Slow down actions by N ms (debugging)
    request_delay=1.0,          # Min seconds between prompts (avoid rate limits)
    use_system_chrome=True,     # Use system Chrome (avoids detection)
    user_data_dir=None,         # Persistent browser profile directory
)

📖 API Reference

GeminiClient (async)

Method	Description
GeminiClient(**kwargs)	Create a new client. See Configuration for all options.
async start()	Start the browser
async close()	Close the browser and save session
async login(session_path=None, login_timeout=120)	Authenticate via Playwright browser window
async login_from_browser(browser_name="chrome", session_path=None)	Authenticate by extracting cookies from your browser
async send_prompt(prompt, *, new_chat=False) → GeminiResponse	Send a prompt and get a response
async new_chat()	Start a new conversation
async set_model(model)	Switch model ("fast", "thinking", "pro")
async get_current_model() → str	Get the currently selected model name
async list_models() → list[str]	List available model names
is_authenticated	bool — Check if logged in

Supports async with context manager.

GeminiClientSync (sync)

Same API as GeminiClient but without async/await. Supports with context manager.

GeminiResponse

Field	Type	Description
text	str	The full response text
prompt	str	The original prompt sent
conversation_id	str | None	Conversation identifier from the URL
model	str | None	Model name (if detectable)
timestamp	datetime	When the response was received

response = client.send_prompt("Hello")
print(response.text)             # "Hi! How can I help you today?"
print(response.prompt)           # "Hello"
print(response.conversation_id)  # "6ea876188b87661c"
print(response.timestamp)        # 2026-05-09 08:43:01
str(response)                    # "Hi! How can I help you today?"

Exceptions

Exception	Description
AuthenticationError	Not logged in or session expired
CaptchaError	CAPTCHA challenge detected
RateLimitError	Google rate limiting detected
ResponseTimeoutError	Response not received within timeout
SelectorNotFoundError	UI element not found (UI may have changed)
BrowserError	Browser launch or runtime error

🔐 Authentication Methods

Method 1: login_from_browser() — Extract cookies from your browser

The easiest way to authenticate. Extracts Google cookies from your regular browser and injects them into the Playwright context.

Requirements:

pip install -e ".[cookies]"

Prerequisites:

• You must be logged into gemini.google.com in your chosen browser
• Your browser must be closed (cookie files are locked while the browser is running)

Usage:

async with GeminiClient(headless=True, session_path="sessions/default.json") as client:
    await client.login_from_browser(browser_name="chrome")
    # Session is saved — next time you can just use login()

Supported browsers: "chrome", "chromium", "firefox", "opera", "edge"

Method 2: login() — Manual sign-in via Playwright window

Opens a Chrome window where you sign in manually. Works in headed mode only.

First time:

async with GeminiClient(headless=False, session_path="sessions/default.json") as client:
    await client.login()  # Sign in manually in the browser window

After that:

async with GeminiClient(headless=True, session_path="sessions/default.json") as client:
    await client.login()  # Uses saved session — no manual sign-in needed

🛡️ Anti-Detection

Google blocks sign-in from browsers it detects as automated. This library uses several strategies to bypass that:

Strategy	How
System Chrome	Uses your installed Chrome instead of Playwright's bundled Chromium
Persistent profile	Real user data directory — looks like a normal browser
Stealth args	--disable-blink-features=AutomationControlled and others
JS patching	Removes navigator.webdriver flag
Realistic UA	Standard Chrome user agent string

If you still get "This browser or app may not be secure"

1. Use login_from_browser() — This bypasses the sign-in entirely by using your existing browser cookies
2. Install Google Chrome (not Chromium) — Google is more lenient with its own browser
3. Use a persistent user data directory — Makes the browser look like a regular user's:

   client = GeminiClient(
       headless=False,
       session_path="sessions/default.json",
       user_data_dir="/home/you/.config/gemini-automation-chrome",
   )
   

🔧 Troubleshooting

Problem	Solution
"This browser or app may not be secure"	Use login_from_browser() instead, or see Anti-Detection
"No Google cookies found in chrome"	Log into gemini.google.com in your browser first, then close the browser
"Not logged in and running in headless mode"	Run with headless=False first, or use login_from_browser()
"CAPTCHA challenge detected"	Run with headless=False to solve it manually
"Rate limit detected"	Increase request_delay (e.g., 3.0) and wait before retrying
"Response did not complete within timeout"	Increase timeout (e.g., 300.0)
"No element found for any selector"	The Gemini UI changed — update selectors in src/gemini_webui/selectors.py

🏗️ How It Works

┌─────────────┐     ┌──────────────┐     ┌─────────────────┐
│  Your Code   │────▶│ GeminiClient │────▶│  Chrome Browser  │
│              │     │              │     │  (System Chrome)  │
│ send_prompt()│     │  ┌────────┐  │     │                  │
│ new_chat()   │     │  │ Auth   │  │     │  gemini.google   │
│ login()      │     │  │Manager │  │     │     .com/app     │
│              │     │  └────────┘  │     │                  │
│              │◀────│  ┌────────┐  │◀────│  Response Text   │
│              │     │  │Response│  │     │                  │
│              │     │  │Parser  │  │     │                  │
│              │     │  └────────┘  │     │                  │
└─────────────┘     └──────────────┘     └─────────────────┘
                           │
                    ┌──────┴──────┐
                    │   Session    │
                    │  Persistence │
                    │ (cookies +   │
                    │  localStorage)│
                    └─────────────┘

1. Browser launch — Starts system Chrome with a persistent profile and stealth arguments
2. Authentication — Either extract cookies from your browser (login_from_browser()) or sign in manually (login())
3. Prompt sending — Types into the contenteditable prompt area and clicks send
4. Response detection — Monitors the "Stop generating" button + content stability to detect when streaming is complete
5. Text extraction — Pulls the response text from the model's response container

⚠️ Limitations

• Text prompts only — No file/image uploads (yet)
• UI changes — Google may update the Gemini UI, breaking selectors. The library uses fallback selector chains for resilience.
• Rate limits — Google may throttle excessive usage. Use request_delay to pace requests.
• Terms of Service — Automating the web UI may violate Google's ToS. Use responsibly.

🛠️ Development

# Install in development mode
pip install -e ".[dev]"

# Run linter
ruff check src/

# Format code
ruff format src/

📄 License

MIT