Skip to content

Getting Started with the App

This document is a tutorial. It shows how to start the app and how to use it.

Install the App

To install the app, obey the instructions in the Installation Guide.

First steps with the App

  1. Create an External Integration. Go to Extensibility → External Integrations → Add. Set the Name and the Remote URL of your LLM endpoint, for example https://ollama.example.com. If the endpoint needs an API key, attach a Secrets Group.
  2. Create an AI Provider. Go to AI Tools → AI Models → AI Providers → Add. Enter a name and select the External Integration from step 1. You can also skip step 1: use the + button beside the field to create the External Integration in a modal, without leaving this page.

  3. Set the Provider type. This is the API dialect that a consuming app uses to address the endpoint. See AI Provider.

    Add an AI Provider Add an AI Provider

  4. Discover the models. Go to Jobs → AI Models → Discover AI Models and run the job. The job reads the model catalog of the provider. It creates one AI Model record for each entry.

  5. Examine the result. Open the detail page of the provider. The AI Models panel on the right lists what the job found.

    An AI Provider after discovery An AI Provider after discovery

What are the next steps?

  • Set the Kind of each model that does embedding. The job creates every model as chat, because the catalog endpoint does not say which is which.
  • Clear the Enabled checkbox on each model that a consumer must not use.
  • Set a default num_predict and temperature on the provider. Override either one on a single model that needs a different value.
  • Put anything else that a call needs in Default parameters on the model.
  • Schedule the Discover AI Models job, to keep the catalog current.
  • Read the catalog from your own code or from the REST API. See External Interactions.

How to register an MCP server

The MCP registry works the same way.

  1. Go to AI Tools → MCP Models → MCP Servers → Add. Give the server a name, select or create its External Integration, and select its transport.
  2. Open the server and select Run Discovery. The job reads what the server advertises and records it.
  3. Examine each new tool. A discovered tool arrives with writable set, because the app assumes that a tool writes until a person has read what it does.

The discovery job needs the optional discovery extra. See External Interactions.

For more examples, read the Use Cases section.