Local / offline backend non‑functional (ClaudeCode backend required, Ollama/Mistral not usable)

[aggressor1523]

The current PentestGPT implementation appears to be tightly coupled to online Claude/OpenAI backends and requires API keys to function. While there are references to “local” or “offline” mode, the local backend is incomplete and not integrated in the execution flow. Attempting to run with GPT4All/Mistral models results in missing classes (e.g., ClaudeCodeBackend) and import errors, indicating hard dependencies on claude_agent_sdk. This prevents use in offline, lab, or air‑gapped environments where pentesting tools are commonly deployed.

Enabling a fully local backend (e.g., GPT4All/Ollama) would significantly increase the tool’s educational and research value, especially for students without access to paid APIs. Decoupling the agent/controller from Claude‑specific classes through a clean backend abstraction layer would greatly improve extensibility. I recommend refactoring backend calls behind a common interface and officially supporting at least one offline model.

Thank you for your excellent work — offline support would make PentestGPT much more broadly usable.

[hetoke]

For people who still can't connect to Ollama server

I use a Modelfile with a content like this

FROM gpt-oss:20b

Then do this to create alias with name openai/gpt-oss-20b

ollama create openai/gpt-oss-20b -f Modelfile

Now with this alias, pentestGPT on my Kali VM will be able to connect to my Ollama server on host