PentestGPT alternatives that accept your own model
PentestGPT alternatives that accept your own model are Strix, Shannon, PentAGI, and Claude Code, Codex or OpenCode.
Legacy PentestGPT only accepts model ids from its own registry, so it cannot name an abliterated model id.
What the registry allows, and what it does not
PentestGPT was published at USENIX Security 2024. Its legacy registry is a fixed map of model ids. (registry.py) A new base URL does not add your model, so this site publishes no patch. The autonomous mode hands the work to Claude Code or Codex. (pentestgpt_agent)
hackingBuddyGPT documents llm.api_base, but llm.py leaves that argument commented out with "TODO: why is this not working anymore." (llm.py) CAI is archived and says "No further releases, bug fixes, security patches or support will be provided." Nebula does not spell out an arbitrary custom endpoint. (Nebula guide)
Which alternative matches the job
Strix
One command, a Docker sandbox, and a base URL.
Set it up White boxSource plus URLShannon
Reads the repository, then tests the running app.
Set it up A teamWeb UIPentAGI
Flows, history and a custom provider block.
Set it up You driveCopilotClaude Code, Codex, OpenCode
The backends PentestGPT's autonomous mode calls.
Set it upKimi K3.1 Pro for judgment, DeepSeek V4.1 Flash for volume, and the API ranking for where they are served.