Connect AI apps ^ 1.12.6

When you use AI on Olares, you typically work with two separate applications: one AI service app provides AI capabilities in the background, and another AI client app provides the chat interface you interact with directly. To make them work together, you must connect them.

This guide explains how to understand your apps, gather the required connection details, and configure the connection.

Learning objectives

By the end of this tutorial, you will be able to:

• Distinguish between AI service apps and AI client apps.
• Understand the essential connection parameters and how to get them.
• Connect common AI apps.

Understand your AI apps

Before configuring a connection, determine which app is providing the AI capabilities and which app is using them. This step ensures you know whether to look for your connection details in the Model Console or in Olares Settings later in this guide.

• AI client apps: They provide the front-end chat interface or workflow canvas you interact with directly, such as LobeHub and Open WebUI. They rely on an AI service app (often called a provider) to perform AI tasks, such as generating text and extracting data.

• AI service apps: They provide AI capabilities for compatible clients over an API, such as chat, search, and speech recognition. Some AI service apps have their own web interface for management, while others run primarily as headless backend services.

  On Olares, AI service apps fall into two categories:

  • LLM service apps: Apps that host large language models (LLMs) for text generation, code completion, and chat. They include eight pre-built model apps, and custom model instances created on Engine Base apps.
  • Other AI service apps: Utility apps that provide non-LLM functions, such as speech recognition (Speaches) and text extraction (PaddleOCR).

Prepare connection details

Connecting an AI client app generally involves configuring up to four key parameters. Gather these details before setting up your client.

Provider and API format LLM services only

In most AI client apps, a provider is the service or vendor that supplies the LLM (such as OpenAI, Anthropic, or Ollama). On Olares, your local LLM service app acts as this provider. Instead of sending requests to a cloud vendor, your client app sends them to your local LLM service app.

Because different providers use different communication rules, they rely on specific API formats. You can think of these formats as the "languages" that the apps use to talk to each other. The two most common formats are OpenAI-Compatible and Ollama.

When configuring an LLM connection, first check which providers your client app supports. Then, in the Model Console, select the API format that matches the provider you chose in the client. The Model Console will dynamically display the correct Base URL according to your selection.

INFO

Non-LLM services like PaddleOCR do not use these generic formats. They communicate using their own tool-specific protocols, so you do not need to configure the provider format for them.

Base URL

The Base URL is the network address (or endpoint) where the AI service app receives and processes your tasks. How you locate it depends on the category of your service app:

• For LLM service apps: Open the app to launch its Model Console. Select the Connection source that matches where your client runs, choose the API format, and then copy the provided Base URL. Copy it exactly as displayed, including any path suffix such as /v1.

• For other AI service apps: Open Olares Settings, go to Applications > [AppName] > Entrances, and then copy the Endpoint URL. Ensure the entrance's Authentication level is set to Internal so other apps can access it without a login barrier.

  Multiple entrances

  Some apps expose more than one entrance. Choose the entrance that matches your client's protocol or use case. For example, use the main entrance for web UI access and a dedicated API entrance for programmatic integrations.

Model name

The model name is the exact identifier of the model. The client sends this ID with every request so the backend service knows which model file to process.

In the Model Console, copy the Model name exactly as displayed.

Copy the full model name

Do not abbreviate the model name or remove any repository prefixes (like unsloth/) or quantization tags (like UD-Q4_K_XL). Otherwise, the client might return an error similar to "Model not found".

API key

An API key (also called an "Auth Token" or "API Token") is a secret credential. The AI client app presents it to the AI service app to prove its identity and gain permission to access the AI capabilities.

For AI service apps deployed locally on Olares, a real API key is usually not required. The internal entrance already trusts requests from other apps in the same cluster.

However, most AI client apps still require a value in the API key field. Enter any placeholder text such as olares or local to proceed.

Examples

The following examples demonstrate how to gather and apply the connection parameters in real client applications. Ensure that the relevant apps are installed before starting.

Connect an LLM service app to LobeHub

In this example, the pre-built model app Gemma 4 26B (Ollama) is the LLM service app, and LobeHub (previously known as LobeChat) is the client app.

1. Open Gemma 4 26B (Ollama) from the Launchpad to launch its Model Console.

2. Ensure that the Model shows READY and the Engine shows RUNNING.

3. Select the connection options that match your client app to get the correct Base URL:

   • Connection source: Select Apps in Olares, because LobeHub is installed directly in the Olares cluster.
   • API format: Select Ollama, because LobeHub's Ollama provider expects this format.

4. Note down the following details:

   • Model name: gemma4:26b
   • Base URL: https://74bfa5ee.alice2026.olares.com

   

5. Open LobeHub, and then go to Settings > AI Service Provider > Ollama.

6. In the Interface proxy address field, paste the Base URL you copied.

7. Ensure Use Client Request Mode is disabled.

   Disable Client Request Mode

   Do not enable Use Client Request Mode in LobeHub. Enabling this forces the application to make frontend browser calls, which can trigger cross-origin (CORS) blocks or Olares security authentication prompts. Keeping it disabled ensures secure, direct backend-to-backend communication.

8. Next to Model List, click Fetch models. The model name gemma4:26b appears in the list.

9. Toggle on gemma4:26b to enable it.

10. On the right of Connectivity Check, select the model name from the drop-down list, and then click Check. When Check Passed appears, the connection is established.

    

Connect PaddleOCR to Open WebUI

In this example, PaddleOCR is the AI service app that provides image text recognition capabilities, and Open WebUI is the client app.

1. Open Olares Settings, and then go to Applications > PaddleOCR > Entrances > PaddleOCR.

   

2. Ensure the Authentication level is set to Internal.

3. Copy the endpoint URL. For example, https://17b4c78a.alice2026.olares.com.

   

4. In Open WebUI, click the profile icon, and then go to Admin Panel > Settings > Documents.

5. In the General section, configure the settings as follows:

   • Content Extraction Engine: Select PaddleOCR-vl.
   • API Base URL: Enter the PaddleOCR endpoint URL you just copied.
   • API Token: Enter any placeholder text. Do not leave this field empty.

   

6. Click Save.

FAQs

How to connect non-AI apps?

The same internal-entrance pattern applies when connecting non-AI apps to each other. For example:

• The *Arrs media stack uses internal entrance URLs to connect Sonarr, Radarr, Prowlarr, Bazarr, and qBittorrent. See Manage your media library with the *Arrs ecosystem.
• SearXNG itself is not an AI model, but it can be connected to an AI client such as Vane for private, enhanced search. See Connect SearXNG to Vane.

Learn more

• Use cases
• Host local large language models with Engine Base apps
• Manage application entrances