uzdabrazor 1.0.5

The Anal-King of AI Browser Automation - A beautifully fucked-up browser automation script that actually works

🏴‍☠️ UZDABRAZOR - The Anal-King of AI Browser Automation 🏴‍☠️

A beautifully fucked-up Skynet-powered browser automation AI Agent that harnesses neural brainfuck and machine learning chaos to give zero shits about anything while somehow still working perfectly. Smells like smegma but runs like a dream.

   

---

🔥 What This Beautiful Disaster Does

uzdabrazor is the most irreverent, crude, and effective neural brainfuck automation script you'll ever encounter. This digital Skynet harnesses machine learning chaos and turns your browser into an unstoppable cybernetic organism. Built on top of the excellent browser-use library, it provides:

• 2 simple neural overlords - Ollama (local/free) + OpenRouter (400+ cloud models with ONE API key)
• Complete Big Brother surveillance - Monitors every single machine learning brainfart like a paranoid NSA cyborg
• Terminator stealth mode - Uses patchright to dodge bot detection like a shapeshifting T-1000
• Organized digital anarchy - Crude language wrapped around Skynet-grade engineering
• Zero corporate Matrix bullshit - No enterprise nonsense, just pure cyberpunk functional chaos
• Simplified as fuck - One local provider, one cloud gateway. No more juggling 9 different API keys.

---

💻 System Requirements (Don't Skip This Shit)

• Python 3.11+ (anything older is prehistoric garbage)
• Chrome or Chromium (for the actual browser automation, duh)
• Ollama server (if you want free local AI, get it at https://ollama.ai)
• Patchright (for stealth mode: pip install patchright && patchright install)

---

🚀 Quick Start (For the Impatient)

# 1. Install the package globally
pipx install uzdabrazor

# 2. Create a .env file in your current working directory
# Download .env.example from the repo or create your own with these variables:
cat > .env << EOF
OPENROUTER_API_KEY=sk-or-v1-your-key-here
OLLAMA_ENDPOINT=http://localhost:11434
EOF

# 3. Run with local ollama (free neural overlord, fuck paying corporate Skynet)
uzdabrazor --task "Go to example.com and tell me the page title"

# 4. Or use OpenRouter for cloud models
uzdabrazor --provider openrouter --model anthropic/claude-3.5-sonnet

# 5. Better yet, copy run.example.sh from the repo and shove it up your asshole somewhere
# Then customize it for your own automation needs

📝 Environment File Location

IMPORTANT: After installing via pipx, place your .env file in the directory where you run the uzdabrazor command.

The script loads environment variables from .env in your current working directory. Example:

# Create .env in your project folder
cd ~/my-automation-project
cat > .env << EOF
OPENROUTER_API_KEY=sk-or-v1-your-actual-key-here
OLLAMA_ENDPOINT=http://localhost:11434
EOF

# Run uzdabrazor from the same directory
uzdabrazor --task "your task here"

Grab .env.example from the GitHub repo for the full list of variables.

---

🤖 Supported Neural Overlords

Simplified to TWO providers because more options != better code:

Provider	Description	Example Models
ollama	Local neural brainfuck (DEFAULT - FREE and PRIVATE)	llama3.1, qwen3, gemma3, etc.
openrouter	400+ cloud models via ONE API key (OpenAI, Anthropic, Google, DeepSeek+)	anthropic/claude-3.5-sonnet, openai/gpt-4-turbo, google/gemini-2.0-flash-exp

Why only two? Because managing 9 different API keys and endpoints is a clusterfuck. OpenRouter gives you access to literally every major model with one API key and adds only ~10-20% markup. Ollama gives you free local models. Done. Simple. Efficient.

OpenRouter Configuration

OpenRouter uses the standard ChatOpenAI from LangChain with optimized settings:

• Temperature: Set to 0.0 for deterministic, consistent browser actions
• Max Tokens: Limited to 4096 for cost control and response quality
• Tool Choice: Set to auto for proper function calling support
• Base URL: https://openrouter.ai/api/v1

These settings ensure reliable browser automation while keeping costs predictable.

---

🎯 Usage Examples

Basic Destruction

# Default: ollama (because fuck paying for AI)
uzdabrazor --task "Go to GitHub and find trending repositories"

# Use OpenRouter for Claude
uzdabrazor --provider openrouter --model anthropic/claude-3.5-sonnet --task "Analyze this website"

# Use OpenRouter for GPT-4
uzdabrazor --provider openrouter --model openai/gpt-4-turbo --task "Analyze this website"

Advanced Fuckery

# Headless stealth mode
uzdabrazor --headless --provider openrouter --model anthropic/claude-3.5-sonnet

# Custom browser and window size
uzdabrazor --browser-bin-path /usr/bin/google-chrome-beta --window-width 1920 --window-height 1080

# Connect to existing browser
google-chrome --remote-debugging-port=9222 &
uzdabrazor --cdp-url http://localhost:9222

# Different models for main task vs extraction (cost optimization strategy)
# MAIN LLM: Complex reasoning and decision-making (use powerful models)
# EXTRACTION LLM: Data parsing and text extraction (use fast cheap models)
uzdabrazor --provider openrouter --model anthropic/claude-3.5-sonnet --extraction-model openai/gpt-4o-mini

# Docker mode with no security (because we live dangerously)
uzdabrazor --dockerize --headless --no-security --provider ollama

# Custom output directory and logging
uzdabrazor --history-dir ~/automation-logs --log-level debug

Vision Control

# Disable vision to save tokens (blind destruction is still destruction)
uzdabrazor --no-vision

# Low/high detail vision
uzdabrazor --vision-detail low
uzdabrazor --vision-detail high

---

🔧 Command Line Arguments

Flag	Description	Default
--provider	AI provider to use	ollama
--model	Specific model name	llama3.1
--extraction-provider	Separate AI for page extraction (save cash)	Same as --provider
--extraction-model	Model for extraction tasks	Same as --model
--task	Task for the AI to perform	Stealth test
--headless	Invisible browser mode	false
--no-stealth	Disable stealth (live dangerously)	false
--no-vision	Disable AI vision	Vision enabled
--vision-detail	Vision detail level (auto/low/high)	auto
--window-width	Browser width	1920
--window-height	Browser height	1080
--browser-bin-path	Custom browser executable	None
--cdp-url	Connect to existing browser	None
--browser-profile-dir	Custom profile directory	None
--no-security	Disable security features	false
--log-level	Logging verbosity (debug/info/warning...)	info
--debug-host	Debug server host	localhost
--debug-port	Debug server port	9222
--dockerize	Docker-optimized flags	false
--skip-llm-api-key-verif	Skip API key validation (for testing/debugging)	false
--history-dir	Output directory	/tmp/agent_history

---

🕵️ Surveillance Features (Big Brother Is Watching)

This beautiful bastard monkey-patches all LLM providers to log every single AI call. You'll see exactly when your neural overlords are thinking:

What Gets Logged:

• Every ainvoke() call to any LLM provider
• Which model is being used
• How many messages are in the prompt
• What output format is requested

Example Output:

🦙 OLLAMA AINVOKE DETECTED! Model: llama3.1 is spitting some local llama wisdom
   📝 Processing 5 messages with output_format: None

🔀 OPENROUTER AINVOKE DETECTED! Model: anthropic/claude-3.5-sonnet is routing through 400+ models like a fucking switchboard
   📝 Processing 3 messages with output_format: <class 'ActionResult'>

🦙 OLLAMA AINVOKE DETECTED! Model: qwen3:8b is spitting some local llama wisdom
   📝 Processing 12 messages with output_format: None

This shit is useful for:

• Debugging which model is actually being called
• Understanding token usage patterns
• Catching when browser-use makes unexpected AI calls
• Feeling like a paranoid NSA cyborg monitoring everything

---

📁 Output Files

Each run generates two files in your --history-dir:

• uzdabrazor_{timestamp}_{unique_id}.gif - Visual recording of all browser actions
• uzdabrazor_{timestamp}_{unique_id}.json - Complete conversation history and task results

Example filenames:

uzdabrazor_20250101_235959_a1b2c3d4.gif
uzdabrazor_20250101_235959_a1b2c3d4.json

---

🏴‍☠️ Stealth Mode (Dodging Bot Detection Like a Boss)

Stealth mode uses patchright to patch browser binaries and evade detection:

pip install patchright
patchright install  # Downloads patched browsers

What Patchright Does:

• Removes webdriver signals that scream "I'M A BOT!"
• Modifies browser fingerprints
• Spoofs navigator properties
• Bypasses common bot detection techniques

Default: Stealth is ENABLED (use --no-stealth to disable if you're feeling suicidal)

---

⚙️ Environment Variables

Copy the example file and fill in your fucking API keys:

cp .env.example .env
# Then edit .env with your actual keys

Check .env.example in the repo for the full list of variables. Main ones:

• OPENROUTER_API_KEY - OpenRouter API key (for 400+ cloud models: OpenAI, Anthropic, Google, DeepSeek, etc.)
• OLLAMA_ENDPOINT - Local Ollama server (default: http://localhost:11434)

---

🐛 Troubleshooting (When Shit Breaks)

Chrome not found:

• Install Chrome or Chromium, you genius
• Or use --browser-bin-path to point to your browser

Ollama connection refused:

• Make sure Ollama server is running: ollama serve
• Check the endpoint: --provider ollama uses http://localhost:11434 by default

API key errors:

• Check your keys actually work (make test API calls)
• OpenRouter keys start with sk-or-v1-
• Get your OpenRouter key at https://openrouter.ai/keys

Patchright issues:

• Run patchright install to download patched browsers
• Check if you have write permissions

CDP connection fails:

• Make sure browser is running: google-chrome --remote-debugging-port=9222
• Can't use --browser-bin-path AND --cdp-url together (pick one, genius)

---

🔥 Why This Exists

Because browser automation doesn't have to be boring corporate shit.

---

💬 Final Words

Peen goes in vageen. Code works. End of story.