Choosing & Managing Your AI

Picking a provider

Vrge routes AI calls through four providers. You choose which one in Settings → AI & Intelligence.

Ollama (Local): runs on your machine, no network, no API key, no cost. Install Ollama separately and point Vrge at whichever model you prefer. First-class for privacy-sensitive work.
Anthropic, ChatGPT, Google Gemini: bring your own API key. Pay the provider directly. Vrge takes no markup and never proxies your content.
Switching to a cloud provider triggers a consent modal with a preview of what the mode in force would send. Nothing leaves your machine without your explicit accept.

Set a per-user daily token budget in Settings → AI & Intelligence. A warning banner appears at 80% usage; a hard stop at 100%. No surprise bills.
A one-toggle kill switch disables all sources, stops scanning, and cancels in-flight AI calls instantly. No confirmation dialog. One click, instant stop.

The AI activity log shows every single call: timestamp, provider, task, token count, estimated cost, and whether content was redacted. Exportable as JSON.
You choose what the cloud sees: with your own cloud key the default is to send the full content to your own provider account. Your key, your data. Turn redaction on and Vrge replaces names, emails, addresses and dollar amounts with tokens before the call, or sends only a locally-written summary (the strict, truncated text when Ollama is off). The “What gets sent” preview shows a sample of what the mode in force for your provider would send, and lists any source set to send differently. Vrge Managed AI is the exception: redaction is floored there, so full content can never be sent through us: the call goes out as a locally-written summary, or as the tokenised, truncated text when Ollama is off.