Made withDifyClip

Dify Quickstart Guide: Build Your First AI Workflow

D

Dify

Creator on the Dify channel

Watch on YouTube ↗

21K viewsSep 19, 20259:47 runtime

Learn to build your first AI workflow application with this Dify quick start guide. This tutorial walks you through creating a 'country infobot' that uses a Large Language Model (LLM) to classify user questions and enhance responses with a fun fact from a custom code block. You…

Chapters

#dify#clip#aillm

Building Your First AI Workflow in Dify

This guide demonstrates how to build a simple "Country InfoBot" using Dify's visual workflow editor. The goal is to classify a user's question, answer it using an LLM, and enhance the response with a fun fact.

Core Workflow Components

The application is built by connecting four main types of nodes:

Start Block: The entry point for the workflow. It receives the user's initial query as a system.query variable, which is then available to all subsequent blocks.
LLM Block: The core of the AI logic. This block sends a prompt to a Large Language Model (like GPT or DeepSeek) to process the user's input.
- Prompting: A custom prompt instructs the LLM on how to behave and what to extract from the user's query.
- Variables: Use {{variable_name}} syntax within the prompt to insert data from previous blocks, like {{start.query}}.
Code Block: Executes custom Python code to add logic or transform data. In this example, it takes a country name as input and returns a hardcoded fun fact.
Answer Block: The final node that aggregates data from previous blocks and formats the final response sent back to the user.

Key Dify Concepts & Techniques

Workflow Orchestration

The visual workspace allows you to connect blocks by dragging from one node's output to another's input. This creates a data flow: Start -> LLM Block -> Code Block -> Answer Block

Structured Outputs

A crucial feature for creating reliable AI applications. By enabling structured outputs in the LLM block, you can force the model to return a consistent JSON object.

You define a schema (e.g., { "question": "string", "country": "string", "answer": "string" }).
The LLM's response is parsed into these fields as variables (llm_block_name.question, etc.).
This makes outputs predictable and machine-readable for downstream use, like feeding the country variable into the Code Block.

Important
When copy-pasting a prompt template with variables, you must manually re-insert the variables using Dify's UI (e.g., by typing {{). Pasted text is treated as a literal string and won't be processed as a variable.

Variable Management

Variables are the data passed between blocks. Dify helps manage this by scoping variables to the block that generates them.

In the Answer Block, you can easily reference outputs from any preceding block, such as:
- {{llm_block_name.answer}} to get the answer from the LLM.
- {{code_block_name.fun_fact}} to get the fun fact from the Python code.

Tip
Giving your blocks descriptive names (e.g., country_llm instead of the default llm-1) makes it much easier to select the correct variables in later steps.