Building Dify Platform Based on Ollama Backend (Part 1): Installation

Purpose: Establish your own workflow through Dify to enhance personal AI capabilities.

Goal: Implement Dify’s Docker container locally to build your own workflow. This article focuses solely on the deployment and installation of Dify.

First, let’s have a look at the origin of the name Dify: Do It For You

Official Dify GitHub address:

https://github.com/langgenius/dify

Now let’s get started.

1. Why Use Personal Platforms Like Dify

For workflow platforms, I personally recommend online platforms like Coze. First, there is no need for local setup, which saves space and setup costs. Second, there is free computing power available for direct use, saving local computing resources, making it quite economical.

However, for enterprises or personal deployments, especially for sensitive or private data, it is not wise to place them on public platforms. Moreover, for researchers, developers, and individual users, it is necessary to build a platform by hand to better understand the construction process of the workflow platform, laying the foundation for future enterprise and team AI platforms.

2. Precautions

First, think carefully about the purpose and costs.

Before building Dify, it is essential to note that the cost of setting up Dify is not low. There are certain requirements for hardware and space, especially the Ollama model and Dify’s knowledge base, which will consume a significant amount of hard disk space.

Hardware and software requirements (must pay attention):

1. Recommended CPU processing capacity is at least 4 cores or more, with a frequency of no less than 2.0GHz. Most current CPUs should meet this requirement.
2. It is advisable to have a larger hard disk since Docker and the knowledge base will take up a lot of space, especially the knowledge base. If handling a lot of content, it can easily exceed tens of GB, so it’s recommended to leave at least 100GB.
3. If running large models locally, the official requirements for local computing power are low. However, due to running models and other issues, it is recommended to have a graphics card with at least 6GB and memory of at least 16GB. If there is no graphics card, only CPU and memory can be used to run smaller models, ideally below 4B, but the performance will definitely be compromised.
4. One last point to note is that it is best not to deploy on Windows Home Edition, as the Windows features in the Home Edition are severely limited, making it difficult to install Docker successfully, and special handling methods may not always work.

3. Practical Steps

The basic sequence of practical steps follows the logic below:

1. Install #git#

2. Install #Ollama#

3. Install #Docker#

4. Install IIS Service

5. Pull #Dify# Source Code

6. Pull #Dify#’s #Docker#

1. Install Git

   #git# can be directly downloaded and installed online.

   Download address: Git – Downloads (git-scm.com)

   https://git-scm.com/downloads

   Simply follow the prompts to install.

2. Install Ollama

   See 【AI Workflow】Ollama Localized Large Model Installation

3. Install 🟢 Docker

   Download from the official website at the link below:

   https://www.docker.com/get-started/

   Choose the corresponding system to download.

   

   After downloading, double-click to install.

   Download installation package address: (link may expire)

   https://desktop.docker.com/win/main/amd64/

   Then double-click to install, the default installation is on the C drive. If it’s Windows Home Edition, it may fail.

   

   Installation completed, note! It will restart automatically!!!!!!

   

   Docker Settings:

   

   Skip login, just skip everything or close and reopen Docker.

4. Restarting Docker will start it automatically.

   
   

   At this point, Docker installation is complete.

   After installation, due to some external images of Docker not being able to be pulled from foreign resources (if you have special methods, that’s another story), you need to change the image to local resources. Be cautious, as this can easily cause Docker to crash. The specific steps are as follows:

   1. Click on Docker settings and open the Docker Engine options.

   2. Copy the content in the code box to a txt file as a backup.

   3. Paste the code below into the Docker text box, this action modifies the source for pulling Docker images.

5. {  "builder": {    "gc": {      "defaultKeepStorage": "20GB",      "enabled": true    }  },  "experimental": false,  "registry-mirrors": [    "https://registry.docker-cn.com",    "https://docker.mirrors.ustc.edu.cn",    "https://registry.docker-cn.com",    "http://hub-mirror.c.163.com",    "https://mirror.baidubce.com",    "https://mirror.ccs.tencentyun.com"  ]}

   Click the Apply & restart button in the bottom right corner to restart the Docker service.

   It should be noted that modifying the Docker source does not necessarily solve the problem completely. If the source is not usable, you need to search online for alternative sources and modify the corresponding websites.

Install IIS Service

The IIS service can be installed on the local Windows system. The installation method for Windows 10 and Windows 11 is slightly different, but not too much. Here, I will take Windows 11 as an example:

Press the Windows key + R to open the Run window, type “control” to open the Control Panel.

In the Control Panel, open “Programs and Features”.

In the Programs and Features interface, select “Turn Windows features on or off” and choose Web Management Tools and World Wide Web Services under Internet Information Services.

It will install automatically after confirmation. Installation is complete.

Since Dify’s default ports are 80 and 8080, there may be conflicts with other web services you have deployed, causing Dify or your personal webpage to fail to open. If you have your own website deployed or port conflicts, see below; otherwise, you can skip this:

1. Open the IIS service.

2. There are two methods to resolve port conflicts:

The first method is to temporarily stop your website service.

The second method is to change the port of your other deployed website.

2.1 First method: Stop your website.

2.2 Change the port.

If the already deployed personal website cannot change the port, simply create a new one and point it to a different port.

Enter a name, physical path, and modify the port. For example, I set the port to 8644; others can also be set, but avoid using ports frequently used by Docker services such as 80, 8080, 3000, 3300, etc.

After confirming, the new website’s address will be 8644. You can access it in the browser by entering localhost:8644 or 127.0.0.1:8644.

In the future, just add the port number at the end of the website address to continue accessing.

Pull Dify source code.

Pull Dify from its GitHub official website.

Create a folder for Dify in a specific directory, for example, I created it in E:\3.CODE\DIFY

In the address bar, type CMD and press Enter (you can also enter this folder directly using cmd or powershell).

Input the following command:

git clone https://github.com/langgenius/dify.git

Then it will start pulling. Note that due to slow access to GitHub from within the country, it often fails to pull, so either use scientific methods or just keep trying. As shown in the figure:

Continue until it successfully reaches 100%.

Pull Dify’s Docker and deploy.

After pulling, navigate to the Dify folder from the previous step, for example, mine is at E:\3.CODE\DIFY\dify.

In the address bar, type CMD.

Input the following command:

docker compose down

This step stops any existing Dify services and uninstalls them; this will also be used for future updates.

Then input:

docker compose pull

This step is crucial as it pulls Dify’s Docker image.

It is common for the pull to fail, and pulling Dify can take a long time, so you must be patient… Moreover, if it stops or fails, any incomplete services will need to be pulled again… Keep waiting until all checks are complete.

After pulling, input the following command to start the Docker service:

docker compose up -d

If successful, you will see Dify in Docker.

Thus, the entire Dify installation is complete.

4. Using Dify

1. Simply open a browser and enter localhost or 127.0.0.1 to access the local Dify homepage.

   

2. The first time you open it, you will be prompted to enter an admin username and password.

   

   Make sure to remember these settings, or you may regret it later.

   Any email will do, just make sure to remember your username and password, which won’t be elaborated on here.

   

3. After setting up, you will be directed to the login screen, where you can input your username and password to log in.

   

4. After logging in, you will enter the Dify page.

   

5. Post-Installation Thoughts and Dify Upgrade (Not Recommended)

To be honest, deploying Dify is quite difficult. However, if you deploy Dify in 2023, it is relatively easier, mainly because the Docker official website is blocked by invisible hands. Fortunately, there are other sources available domestically as alternatives, but the speed and quality are really unsatisfactory and unstable, requiring constant adjustments to the source.

Therefore, unless there are significant upgrades, it is not recommended to upgrade Dify’s Docker image after pulling it once; it is too difficult.

However, if necessary, to facilitate upgrades, you can create an update.bat file in the previously installed Dify directory for convenience:

The specific method is to create a new txt document, double-click to edit it, and enter the following code:

cd dify/docker
git pull origin main
docker compose down
docker compose pull
docker compose up -d
pause

Save the document and rename it with a .bat extension, whether as update.bat or Upgrade.bat, but the extension must be .bat.

In the future, simply double-click this bat file to automatically stop services, pull new images, and deploy services.

Please note that unless necessary, avoid upgrading, as pulling a new image can easily lead to issues, and the failure rate can be high in some cases.