The AI runs on your machine PRO

Charon's AI driver — the one that crawls, exploits and writes the engagement report — talks to a local model. Your traffic, credentials and findings never leave your machine, which is what makes it usable on a real engagement.

ModelCharon AI (9B, Q6_K)
Size6.85 GB
FormatGGUF (llama.cpp)
Runs onLM Studio
Needs16 GB RAM minimum

Download with a Pro key

Setting it up

Three steps: install LM Studio, drop the model where it looks for it, then point Charon at it.

1. Install LM Studio

Get it from lmstudio.ai (free, Windows and macOS). Open it once so it creates its folders, then close it.

2. Put the model where LM Studio looks

LM Studio expects <models>/<publisher>/<model>/file.gguf — two folder levels, or it won't appear in the list.

macOS

~/.lmstudio/models/charon/charon-ai/Charon-AI-Q6_K.gguf

Windows

%USERPROFILE%\.lmstudio\models\charon\charon-ai\Charon-AI-Q6_K.gguf

If you changed LM Studio's models directory, use that path instead — it's shown under My Models in the app. The two folder levels (charon/charon-ai) still apply.

3. Start the local server

  1. Open LM Studio and go to Developer (the >_ tab).
  2. Load Charon AI (9B, Q6_K) from the model dropdown.
  3. Turn the server on. It listens on http://localhost:1234.
  4. Leave the context length at its default unless you have RAM to spare — a longer context uses more memory.

4. Point Charon at it

In Charon, open Settings → AI and enter:

Provider    OpenAI-compatible
Base URL    http://localhost:1234
Model       charon-ai
API key     (leave blank)

Save, then toggle AI on. The model name must match what LM Studio shows in its dropdown. The base URL is the host only — Charon appends the rest of the path itself, so don't include /v1.

You're set: the AI tab now has Watch and Auto, the engagement log, and the report.

Getting the most out of it

Memory. The model holds about 7.4 GB while loaded. On a 16 GB machine, running it at the same time as a thorough scan and a browser will push the machine into swap and slow everything down. Charon releases the model when you close a project and reloads it on demand, so you rarely need to think about it — but if a scan feels slow, eject the model in LM Studio for the duration.

What the AI needs the model for. The exploitation chains are deterministic and run without it. The model is what decides the next move, explains a finding, and writes the report.