Self-hosted server

AI on your server

AI is off by default in every image. For what users can do with AI, see AI assistant.

Choose where AI runs#

  1. Sign in as an administrator and open Settings → Server → AI Engine.
  2. Under Connection, choose an option.
  3. Select Save Changes, then Restart Now. For your own engine on latest-fat, restart the container instead, because Restart Now does not start the bundled engine.

AI connection settings with Off, Run your own engine, and Use Stirling Cloud AI

Option What it means Set it up
Off No AI tools anywhere in the app, and nothing leaves this server. Default
Run your own engine An AI engine you host, using a model provider you choose, such as Claude, GPT or a local model. You hold the keys and pay the provider. Run the AI engine
Use Stirling Cloud AI No engine, provider key or model choice. AI runs on Stirling Cloud and is billed to the Stirling account this server is linked to. Use Stirling Cloud AI from your server

Without an Enterprise licence, AI tools count against the server's monthly processing allowance, even with your own engine and provider key.

For what each option sends off the server, see AI security.

The settings page needs login enabled; otherwise set aiEngine.enabled and aiEngine.mode in settings.yml:

yaml
aiEngine:
  enabled: true
  mode: SELF_HOSTED
bash
AIENGINE_ENABLED=true
AIENGINE_MODE=SELF_HOSTED

Turn capabilities on or off#

All six capabilities are on by default. To turn one off, switch it off under Capabilities on the same page, then save and restart. Capabilities that are off are hidden in the app.

Capability Setting
Chat assistant aiEngine.features.chat
Document questions aiEngine.features.documentQuestions
Create PDF from prompt aiEngine.features.createPdf
Math auditor aiEngine.features.mathAuditor
PDF comment agent aiEngine.features.pdfComment
Document classification aiEngine.features.classify

For example, to turn off PDF creation:

yaml
aiEngine:
  features:
    createPdf: false
bash
AIENGINE_FEATURES_CREATEPDF=false
  • Conversation stays available while either Chat assistant or Document questions is on. Turn both off to disable it.
  • Processor Classification needs Document classification. To route documents without AI, use Routing rules on document properties.
  • Create PDF from prompt needs WeasyPrint on the Stirling PDF server. The standard and fat Docker images include it. On a JAR install, put weasyprint on the PATH or set system.customPaths.operations.weasyprint to its location.
  • For AI tools used through MCP clients, see MCP server settings.