32dots HEIDELBERG AI
Session 2 easy

Swap the model and publish your app

USE 0 - 10 min

Change the model behind your app without rebuilding it

Dify is model-provider agnostic: you can swap between OpenAI, Anthropic, Groq, local Ollama/LM Studio and dozens more through one UI without changing the app. The app you built — its knowledge base, instructions, and chat URL — stays exactly the same; only the model doing the thinking changes. This lesson is where the GUI-builder design pays off.

  1. 1 Open your knowledge chatbot from the previous lesson in the Studio.
  2. 2 Find the model selector for the app. Note which provider/model is currently set.
  3. 3 Switch to a different provider — for example from a hosted model to a local one served by Ollama or LM Studio, or between two cloud providers. (You configure providers once in Dify's settings; then any app can use them.)
  4. 4 Re-ask the same grounded question you used before.
  5. 5 Compare the new answer to the old one — same knowledge base, different model.

The same app answered the same question on two different models, with no change to its knowledge base or instructions.

UNDERSTAND 10 - 18 min

Why your apps are callable from your own code

Everything you have built in the GUI is also reachable programmatically. Dify exposes a Service API (its own chat-messages and completion-messages endpoints), so every app you build is callable from your own code with the app's API key — but the app itself is still BUILT in the GUI, not via the API.

Key concept

Dify separates building from calling. You assemble the app visually — knowledge base, model, instructions — and publish it as a public chat URL or an embeddable widget for humans, AND as a Service API endpoint for your own scripts (Dify's own chat-messages or completion-messages endpoint, called with the app's API key). That means a chatbot you built by drag-and-drop can be invoked from Python or any HTTP client — and it carries your knowledge base and instructions with it.

  1. ?When you swapped models, did the citations stay grounded? Why does the knowledge base survive a model change?
  2. ?Which delivery surface fits your use case best — a public chat URL, an embedded widget, or the Service API?
  3. ?If you self-host Dify under the Dify Open Source License, who controls which models and data the app can reach?
BUILD 18 - 25 min

Pick the model that fits your constraint

Model choice is a trade-off between quality, speed, cost, and privacy — and Dify makes it a dropdown, so you can decide with evidence instead of guesswork.

Run the same grounded question on two models and choose one as your app's default, with a stated reason.

  1. 1 Pick two models from different providers configured in your Dify (e.g. one cloud, one local).
  2. 2 Ask each the same question against your knowledge base.
  3. 3 Compare answer quality, speed, and what left your machine (a local model keeps data on-prem).
  4. 4 Set the winner as the app's model and write one sentence on why.
Deliverable

Your app set to a chosen model, plus a one-sentence justification (quality / speed / cost / privacy).