AI Assistant
Overview
Nemeton comes with an AI assistant. In the app it is called "AI Agent". You open it with the AI Agent button in the toolbar; the agent window shows the progress of a task. The assistant controls your VMs with the tools of the MCP interface and is subject to the same locks as an MCP client.
The chat permanently shows a notice that you are chatting with an AI.
Providers
Choose the provider under Settings → AI Agent → Provider:
| Provider | What you need |
|---|---|
| Nemeton Cloud | A Nemeton Cloud subscription. You take it out in the web store, with the customer account Nemeton is signed in with. |
| Claude (Anthropic), OpenAI, Google Gemini | Your own API key. You pay the provider directly. |
| Ollama (local) | No subscription and no key. Ollama runs free of charge on this Mac. |
Model and endpoint can be changed on the same page.
Consent and Setup
Before its first use, the assistant asks for your consent. Without it, nothing is sent to an AI service. If you withdraw your consent, the assistant is blocked immediately.
If something is still missing, such as consent, sign-in, the subscription or a key, the chat names the next step and takes you there with one click.
Confirmation Mode
Under Settings → AI Agent → Security you decide when the assistant asks first. The default is "Safe".
| Mode | Behavior |
|---|---|
| Safe | You confirm destructive actions first. |
| Paranoid | Only read tools run automatically, everything else asks first. |
| Watch | Like Paranoid; only the reason given in the prompt differs. |
| Autonomous | Confirmation only for access to other VMs. |
- You confirm access to any VM other than the active one in every mode.
- Before the assistant reads or writes files on this Mac, it asks. In "Autonomous" mode it refuses such access.
- The locks under API & MCP also apply to the assistant: it refuses calls to a VM whose remote control is locked, even read-only ones. If destructive actions are switched off there, it refuses them instead of asking.