Skip to main content
Dify is an open-source LLM application development platform for building chatbots, agents, and workflows visually. Dify can connect to the ScitiX Model Inference API through the OpenAI-API-compatible provider plugin, so Chat, Agent, and Workflow apps can use ScitiX models.

Configure

Prepare a Dify workspace first. You can use Dify Cloud or a self-hosted deployment. The plugin marketplace requires Dify 1.x.
Only workspace owners and admins can manage model providers. The API Key is shared across the workspace.

1. Install the provider plugin

  1. Open Integrations → Model Provider from the left sidebar.
  2. In the Install model providers list, find OpenAI-API-compatible published by langgenius, then click Install. If OpenAI-API-compatible does not appear on the first page, use the search box or click Discover more in Dify Marketplace.

2. Add a ScitiX model

The example below adds glm-5.2.
  1. On the OpenAI-API-compatible provider card, click Add Model, fill in the required fields, and save. Keep the following in mind:
    • You can keep the Streaming and vision settings at their defaults.
    • The same plugin supports LLM, Text Embedding, Rerank, Speech2Text, and TTS model types. If your app needs these capabilities, choose the corresponding models from the model catalog.
    • Agent apps require a model with native tool calling support. glm-5.2 supports this capability. If you switch models, check the platform’s tool-calling model list first.
    • If saving fails, or if the model errors when first used in an app, check that the endpoint URL ends with /v1 but not /, and make sure the API Key has no extra spaces.
  2. Set glm-5.2 as the workspace default model. In the top-right corner of the Model Provider page, click Default Models. Before you configure the default model, Dify shows a “Configure a default system model” prompt. After the default model is set, apps can use it automatically.

Verify

Create an app, such as a Chat app in Studio or an agent in the Agents builder. Confirm that the app uses glm-5.2, then send a message. A normal streaming reply confirms that the connection works.