Skip to main content
PentAGI supports Google’s Gemini models through the Google AI API, offering state-of-the-art reasoning capabilities, multimodal features, and massive context windows.

​
Environment Variables

​
GEMINI_API_KEY
string
required
Your Google AI API key. Obtain it from Google AI Studio.
​
GEMINI_SERVER_URL
string
default:"https://generativelanguage.googleapis.com"
Google AI API endpoint URL. Use the default for standard Google AI access.
​
PROXY_URL
string
Optional HTTP proxy URL for network isolation (e.g., http://your-proxy:8080).

​
Configuration Example

Add the following to your .env file: 
# Basic Gemini setup
GEMINI_API_KEY=AIza...
GEMINI_SERVER_URL=https://generativelanguage.googleapis.com

# Using with proxy
GEMINI_API_KEY=AIza...
PROXY_URL=http://your-proxy:8080

​
Supported Models

Gemini provider offers advanced features across multiple model families:

​
Gemini 2.5 Series (Thinking Models)

Advanced reasoning models with step-by-step analysis:
  • gemini-2.5-flash-thinking - Fast thinking model with extended reasoning
  • gemini-2.5-pro-thinking - Advanced reasoning for complex security analysis

​
Gemini 2.5 Series (Standard)

High-performance models:
  • gemini-2.5-flash - Ultra-fast inference for real-time scanning
  • gemini-2.5-pro - Flagship model for comprehensive security assessments

​
Gemini 2.0 Series

Next-generation experimental models:
  • gemini-2.0-flash-exp - Cutting-edge experimental model
  • gemini-2.0-flash-thinking-exp - Experimental thinking capabilities

​
Gemini 1.5 Series

Proven models with large context:
  • gemini-1.5-flash - Fast and cost-effective
  • gemini-1.5-flash-8b - Smaller, efficient variant
  • gemini-1.5-pro - Powerful multimodal analysis

​
Features

  • Thinking Capabilities: Advanced reasoning models with step-by-step analysis
  • Multimodal Support: Text and image processing for comprehensive assessments
  • Large Context Windows: Up to 2M tokens for analyzing extensive codebases
  • Cost-Effective Options: From high-performance pro models to economical flash variants
  • Real-time Performance: Ultra-fast flash models for vulnerability scanning

​
Context Windows

Model FamilyContext WindowBest For
Gemini 2.5 Pro2M tokensAnalyzing entire codebases and documentation
Gemini 2.5 Flash1M tokensLarge-scale security assessments
Gemini 1.5 Pro2M tokensComprehensive multimodal analysis
Gemini 1.5 Flash1M tokensCost-effective large context
Gemini’s massive context windows make it ideal for:
  • Analyzing complete application source code
  • Processing extensive security documentation
  • Reviewing multiple vulnerability reports simultaneously
  • Comprehensive threat modeling across large systems

​
Model Selection by Agent Type

PentAGI automatically selects appropriate Gemini models based on requirements:
Agent TypeRecommended ModelUse Case
Primary Agentgemini-2.5-pro-thinkingCore penetration testing with reasoning
Assistantgemini-2.5-flash-thinkingMulti-step workflows with fast thinking
Generatorgemini-2.5-proReport and exploit generation
Searchergemini-2.5-flashFast information gathering
Codergemini-2.5-proComplex exploit development
Pentestergemini-2.5-flash-thinkingReal-time penetration testing

​
Advanced Configuration

Customize Gemini models using YAML configuration: 
primary_agent:
  model: "gemini-2.5-pro-thinking"
  temperature: 0.7
  top_p: 0.95
  top_k: 40
  max_tokens: 8192
  # Gemini-specific parameters
  safety_settings:
    - category: "HARM_CATEGORY_DANGEROUS_CONTENT"
      threshold: "BLOCK_NONE"

​
Parameters

​
temperature
float
default:"0.7"
Controls randomness (0.0-2.0). Lower values make output more deterministic.
​
top_p
float
default:"0.95"
Nucleus sampling parameter (0.0-1.0). Controls diversity of output.
​
top_k
integer
default:"40"
Top-k sampling parameter. Limits vocabulary for each step.
​
max_tokens
integer
Maximum number of tokens to generate.
​
safety_settings
array
Configure safety filters. For security testing, you may need to adjust thresholds.
When performing security testing, you may need to adjust Gemini’s safety settings to prevent false positives blocking legitimate security research content. Set appropriate thresholds in your configuration:
safety_settings:
  - category: "HARM_CATEGORY_DANGEROUS_CONTENT"
    threshold: "BLOCK_NONE"
  - category: "HARM_CATEGORY_HARASSMENT"
    threshold: "BLOCK_ONLY_HIGH"

​
Pricing

Gemini offers competitive pricing with generous free tier:
ModelFree TierPaid Tier
Gemini 2.5 Flash15 requests/minHigher limits
Gemini 2.5 Pro2 requests/minHigher limits
Gemini 1.5 Flash15 requests/minPay per token
Gemini 1.5 Pro2 requests/minPay per token
Check Google AI Pricing for current rates and limits.
The free tier is suitable for development and small-scale testing. For production penetration testing, consider upgrading to paid tier for higher rate limits.

​
Vertex AI Alternative

For enterprise deployments, you can access Gemini models through Google Cloud Vertex AI:
  • Enhanced security with Google Cloud IAM
  • VPC integration for private networking
  • Enterprise SLAs and support
  • Custom model tuning options
See Vertex AI documentation for details.

​
Troubleshooting

​
Authentication Errors

If you see authentication errors:
  1. Verify your API key at Google AI Studio
  2. Ensure the API key is enabled for the Gemini API
  3. Check for typos in the .env file

​
Rate Limits

Google AI enforces different rate limits:
  • Free tier: 15 requests/min for flash models, 2 requests/min for pro models
  • Paid tier: Higher limits based on quota allocation
Monitor usage in Google AI Studio and request quota increases if needed.

​
Safety Filters Blocking Content

If legitimate security research is blocked:
  1. Adjust safety settings in your configuration (see Advanced Configuration)
  2. Rephrase prompts to be more neutral
  3. Use thinking models which have better context understanding

​
Model Availability

Some models require:
  • Early access program enrollment for experimental models
  • Specific API versions
  • Geographic restrictions may apply
Check the Google AI documentation for current model availability.

Build docs developers (and LLMs) love

Get started for freeTalk to us