[anonrouterai]

Summary

• Adds AnonRouter to the maintained pentestgpt_legacy provider layer using its existing OpenAI-compatible client.
• Adds the dynamic model form anonrouter:<provider>/<model> and passes the nested model ID unchanged to https://api.anonrouter.ai/v1.
• Reads a dedicated ANONROUTER_API_KEY, with optional ANONROUTER_BASE_URL override, and keeps provider credentials isolated.
• Adds CLI/configuration guidance and focused registry, configuration, and factory tests.

Why

I maintain AnonRouter and implemented this through PentestGPT's existing dynamic-provider pattern. A static model list would quickly drift, so the prefix accepts any exact model ID enabled for the user's AnonRouter account. This also avoids adding a dependency or duplicating the OpenAI-compatible transport.

The change is scoped to pentestgpt_legacy, whose native LLM layer supports HTTP providers. pentestgpt_agent is intentionally untouched because its provider boundary is the external unified-agent Claude/Codex CLI adapter.

Usage

export ANONROUTER_API_KEY='<key>'
pentestgpt-legacy \
  --reasoning-model anonrouter:<provider>/<model> \
  --parsing-model anonrouter:<provider>/<model>

Checks

• Root tests excluding Docker — 141 passed
• Legacy focused suite — 33 passed
• Ruff check and format — passed
• Strict mypy for the touched runtime — passed
• uv lock --check — passed
• Root and nested-agent package builds — passed
• Nested-agent suite — 118 passed, 1 skipped
• Live smoke through PentestGPT's actual get_client/AsyncOpenAI path — successful authenticated completion with openai/gpt-4o-mini