PowerToolBox
Download

Configure AI resolutions

The findings grid has an AI Resolution action per finding and a Generate All button that fills in every finding that does not have a resolution yet. Out of the box these run in demo mode: answers come from local templates with a simulated delay, no network calls are made, and nothing leaves the machine. Treat demo answers as canned guidance, not as analysis of your flow.

Point the app at a real provider

There is no settings UI for AI in 1.2.0; the settings file is the only way.

  1. Close PowerToolBox.
  2. Open %APPDATA%\PowerToolBox\settings.json in a text editor. Run the app once first if the file does not exist yet.
  3. Set the AI keys:
{
  "AiProvider": "OpenAI",
  "AiApiKey": "paste-your-key",
  "AiEndpoint": "https://api.openai.com",
  "AiModel": "gpt-4o"
}
  1. Save the file and restart the app.
  • AiProvider — "OpenAI" or "Azure OpenAI". Anything else, including the default "None", keeps demo mode.
  • AiApiKey — your API key. It is stored DPAPI-protected: you can paste plaintext, and the app writes it back encrypted for your Windows user on the next save.
  • AiEndpoint — the base URL requests go to. For Azure OpenAI, use your resource endpoint.
  • AiModel — the model name (OpenAI) or deployment name (Azure OpenAI). OpenAI falls back to gpt-4o when empty; Azure OpenAI requires a value.

You can also edit ResolutionPrompt, the system prompt used when resolutions are generated.

Keys that exist but do nothing yet

The settings file also contains AiEnabled, the per-feature toggles (AiUseForResolution, AiUseForExecutiveSummary, AiUseForBestPractices), and AiTimeoutSeconds / AiMaxTokens. Version 1.2.0 does not read them: the provider selection alone switches between demo and live mode, and requests use a fixed 60-second timeout.

What gets sent

With a real provider configured, a resolution request contains finding metadata only:

  • rule ID, title, severity, category, description, and the remediation text.

Raw evidence and flow definitions are never included. Requests go only to the endpoint you configured, authenticated with your key. The F1 assistant panel uses the same backend, so a configured provider upgrades it from demo answers as well — its greeting message shows which mode is active.

Errors from a live provider surface when you generate a resolution, with details in the logs; see Logs and diagnostics. The full settings schema is in the Configuration reference.

Last updated: July 27, 2026