[Enhancement]: DuckDuckGo Search and Ollama

[samiux]

Target Component

External Integrations (LLM/Search APIs)

Enhancement Description

Is there a way to replace Google API with DuckDuckGo? Meanwhile, is it possible to use Ollama API instead of online LLM API?

Technical Details

No response

Designs and Mockups

No response

Alternative Solutions

No response

Verification

• I have checked that this enhancement hasn't been already proposed
• This enhancement aligns with PentAGI's goal of autonomous penetration testing
• I have considered the security implications of this enhancement
• I have provided clear use cases and benefits

[asdek]

Hello @samiux

Thank you for your interest in enhancing PentAGI

1. Replacing Google API with DuckDuckGo:

Yes, you can replace the Google API with DuckDuckGo. We have recently prepared changes in PR#20 that add support for DuckDuckGo and Perplexity APIs as search engines within PentAGI.

Now DuckDuckGo is available without any setting out of the box.

2. Using Ollama API instead of online LLM API:

Support for Ollama was present in the early stages of the project. However, we were unable to find models that effectively solved tasks within our attack simulation framework, so support for Ollama was removed. We believe that reintroducing Ollama support in the future is feasible, especially as models continue to improve.

Currently, our main development focus is on researching the capabilities of Large Language Models (LLMs) within AI agents. It appears that only the more powerful LLMs are capable of handling the complex tasks in our simulations. As models become more advanced, integrating options like Ollama may become more practical.

Enhancements in PR#20:

In addition to adding support for DuckDuckGo and Perplexity, PR#20 introduces the ftester utility. This tool allows you to test various functions directly within the running PentAGI container, including agent logic. It's a powerful resource for development and debugging, enabling you to:

• Directly test individual components without running the entire system
• Debug AI agent behaviors in specific flow contexts
• Verify environment variables and external service connectivity
• Develop and refine new prompts and agent behaviors with quick feedback

For detailed information about the ftester utility and other changes, please see the PR description and the updated documentation.

Exploring Provider Configurations and Test Results:

We encourage you to review the configurations for different providers and the results of our tests to gain insight into how various models perform within PentAGI. Here are some resources that might be helpful:

• Provider Configuration Files:

  • openrouter.provider.yml
  • deepinfra.provider.yml
  • deepseek.provider.yml

• Test Reports:

  • openrouter-report.md
  • deepinfra-report.md
  • deepseek-report.md

• Testing Instructions:

  • Testing LLM Agents

These resources provide examples and insights into the viability of different models within PentAGI's framework. They can guide you in configuring alternative providers and understanding their performance.

If you have any further questions or need assistance with the configuration, please don't hesitate to ask.

[samiux]

Thank you very much.