Skip to main content
PentAGI integrates with Anthropic’s Claude models, known for their exceptional safety, reasoning capabilities, and sophisticated understanding of complex security contexts.

​
Environment Variables

​
ANTHROPIC_API_KEY
string
required
Your Anthropic API key. Obtain it from Anthropic Console.
​
ANTHROPIC_SERVER_URL
string
default:"https://api.anthropic.com/v1"
Anthropic API endpoint URL. Use the default for standard Anthropic API access.
​
PROXY_URL
string
Optional HTTP proxy URL for secure environments and network isolation (e.g., http://your-proxy:8080).

​
Configuration Example

Add the following to your .env file: 
# Basic Anthropic setup
ANTHROPIC_API_KEY=sk-ant-...
ANTHROPIC_SERVER_URL=https://api.anthropic.com/v1

# Using with proxy for secure environments
ANTHROPIC_API_KEY=sk-ant-...
PROXY_URL=http://your-proxy:8080

​
Supported Models

Anthropic provider delivers superior capabilities across multiple Claude model families:

​
Claude 4 Series

Latest flagship models with exceptional reasoning:
  • claude-4-sonnet - Advanced reasoning for sophisticated penetration testing
  • claude-4-opus - Most capable Claude model for complex security research

​
Claude 3.7 Series

Extended thinking capabilities:
  • claude-3.7-sonnet - Step-by-step thinking for methodical security research

​
Claude 3.5 Series

High-performance models:
  • claude-3-5-sonnet - Balanced performance for comprehensive security analysis
  • claude-3-5-haiku - Blazing-fast model for real-time vulnerability scanning

​
Claude 3 Series

Proven models for various use cases:
  • claude-3-opus - Powerful model for deep security analysis
  • claude-3-sonnet - Versatile model for threat hunting
  • claude-3-haiku - Fast model for high-volume scanning

​
Features

  • Advanced Reasoning: Claude 4 series with exceptional reasoning capabilities
  • Extended Thinking: Claude 3.7 with step-by-step analytical thinking
  • High-Speed Performance: Haiku models for blazing-fast vulnerability scans
  • Comprehensive Analysis: Sonnet models for complex security analysis
  • Safety-First Design: Built-in safety mechanisms for responsible security testing
  • Large Context Windows: Up to 200K tokens for analyzing extensive codebases

​
Model Selection by Agent Type

PentAGI leverages Claude’s advanced understanding of security contexts:
Agent TypeRecommended ModelContext WindowUse Case
Primary Agentclaude-4-sonnet200K tokensCore penetration testing workflows
Assistantclaude-3.7-sonnet200K tokensMulti-step security analysis
Generatorclaude-4-sonnet200K tokensExploit and report generation
Searcherclaude-3-5-haiku200K tokensFast information gathering
Coderclaude-4-opus200K tokensComplex exploit development
Pentesterclaude-3-5-sonnet200K tokensDedicated penetration testing
Claude models are particularly effective for:
  • Complex security analysis requiring deep reasoning
  • Analyzing large codebases and documentation
  • Generating detailed penetration testing reports
  • Ethical security testing with built-in safety guardrails

​
Pricing

Anthropic uses token-based pricing. Check Anthropic Pricing for current rates. Example pricing tiers:
ModelPerformanceRelative Cost
Claude 3 HaikuFastestLowest
Claude 3.5 HaikuVery FastLow
Claude 3 SonnetBalancedMedium
Claude 3.5 SonnetHighMedium-High
Claude 3.7 SonnetExtended ThinkingMedium-High
Claude 4 SonnetAdvancedHigh
Claude 3/4 OpusMost CapableHighest
Claude models have different rate limits based on your API tier. Monitor your usage in the Anthropic Console to avoid interruptions.

​
AWS Bedrock Alternative

You can also access Claude models through AWS Bedrock for enterprise deployments:
  • Enhanced security with AWS IAM integration
  • VPC endpoint support for private networking
  • Enterprise compliance certifications
  • Pay-as-you-go or provisioned throughput options
See AWS Bedrock documentation for configuration details.

​
Advanced Configuration

Claude models support various parameters for fine-tuning: 
assistant:
  model: "claude-4-sonnet"
  temperature: 0.7
  top_p: 0.9
  max_tokens: 8192
  # Claude-specific parameters
  stop_sequences: ["\n\nHuman:", "\n\nAssistant:"]

​
Parameters

​
temperature
float
default:"0.7"
Controls randomness (0.0-1.0). Lower values make output more focused and deterministic.
​
top_p
float
default:"0.9"
Nucleus sampling parameter (0.0-1.0). Controls diversity of output.
​
max_tokens
integer
Maximum number of tokens to generate. Claude models support up to 4096 output tokens.
​
stop_sequences
array
Custom sequences where the model should stop generating.

​
Troubleshooting

​
Authentication Errors

If you see authentication errors:
  1. Verify your API key at Anthropic Console
  2. Ensure your account has active credits or a valid payment method
  3. Check for typos in the .env file

​
Rate Limits

Anthropic enforces different rate limits based on your usage tier:
  • Free tier: Limited requests per minute
  • Build tier: Increased limits for development
  • Scale tier: Production-ready rate limits
Check your tier and limits in the Anthropic Console.

​
Context Length Errors

If you exceed the context window:
  • Enable PentAGI’s automatic context summarization
  • Split large analyses into smaller chunks
  • Use models with larger context windows (all Claude models support 200K tokens)

​
Model Availability

Some Claude models may require:
  • API access approval for newer models (like Claude 4)
  • Specific API version headers
  • Beta program enrollment
Check the Anthropic documentation for current model availability.

Build docs developers (and LLMs) love

Get started for freeTalk to us