Step-by-Step Guide: Connect AI Large Language Models to Xiaomi Smart Speaker for Free - No Coding Required, Beginner-Friendly Detailed Tutorial!

Supported Xiaomi speaker device models

✅ Runs perfectly

? Runs normally

Some models' MIoT interface doesn't support querying device playback status or has abnormal status queries, such as the Xiaomi Speaker Play Enhanced Edition (L05C), which will cause some MiGPT functions to be abnormal and unable to use continuous conversation, etc. In this case, you need to disable streamResponse.

❌ Not supported

Example devices: I'm using the Xiaomi AI Speaker Gen 1, S12

• First, just look at the model number and check if your Xiaomi speaker model is in the list above

Local Deployment Tutorial (Node.js)

• This solution requires your computer to keep the migpt service running at all times. Once you turn off the computer, the large language model stops working. You can think of Xiaoai as your computer speaker, but one that can answer questions anytime - actually pretty interesting

Preparation complete, deployment begins:

Step 1: Software and environment configuration needed on your computer:

• Python 3.8 or above (https://www.python.org/)

• Vscode （ https://code.visualstudio.com/）

• nodeJS （https://nodejs.org/zh-cn）

• Github account required (https://github.com/)

• GitHub is often inaccessible. You can download Steam++, https://steampp.net/

• For installation instructions, refer to Teacher Dayu's tutorial: Install VSCode, NodeJS, Git, etc. on your local computer

1. Search for windows powershell

1. Run as administrator to enter the command line tool page

1. Enter the following command and press Enter

• In this step, we're installing software and configuring the basic environment

Step 2: Prepare the Xiaomi speaker

1. Check the model of your Xiaomi speaker

2. Model 1: Xiaomi AI Speaker (1st Generation), according to the table mentioned above, the model code is S12 (if the bottom label is inconsistent, please refer to the model code in the table)

1. Model 2: Xiaomi AI Speaker (2nd Generation), model code is L15A (check the bottom label)

1. Download the Mi Home APP or Xiaoai Speaker APP on your phone. Here are a few parameters to remember:

• Xiaomi Account: After logging into Mi Home or Xiaoai Speaker APP, you can see the ID in the "My" interface. Mine here is a string of numbers starting with 9

• Remember the password for logging into your Xiaomi account: Make sure to log in using the password, as you'll need it for configuration later

• Configure the speaker device through Mi Home APP or Xiaoai Speaker APP, connect it to your home network. Please ensure the network you're connected to is on the same LAN as your computer

• Remember the speaker's name: Mine here is Xiaomi AI Speaker

• At this step, the parameters we need are: Xiaomi UID, login password, and the speaker's bound name. The name I bound here is Xiaomi AI Speaker

Step 3: Apply for the large model's API interface, using Silicon as an example

• Official tutorial documentation for API integration:

• Taking the silicon API I typically use as an example, there are numerous open-source models (Yi, Qwen, Llama, Gemma, etc.) available for free. It also comes with a 14 yuan trial credit with an unknown expiration date - in my opinion, this is the most convenient and affordable API interface out there

• silicon registration and usage URL:

• After registering and logging in, click on API Keys in the left sidebar, then click Create New API Key

• silicon supports various large language models, as well as text-to-image, image-to-image, and text-to-video generation - feel free to explore on your own

• At this step, once you have the silicon key, we can call the Qwen 2.5 model, which is perfectly adequate for daily conversations and is completely free to use

• In this step, we registered a silicon account, applied for an API key, and saved the key - we'll need to enter it in the configuration later

Step 4: Go to GitHub and copy the mi-gpt project

• Open the official repository: https://github.com/idootop/mi-gpt

• Fork the project to your own account

1. Click Fork

1. Click Create fork in the bottom right corner

1. The project has now been forked to your own repository

1. How to find the project in your repository

1. Copy the repository address: first confirm you're on your own repository project page, then click the green "code" button, then click the copy button in the address bar

• This step copies the official project to your own repository and gets your repository address

Step 5: Open vscode, clone the project, and locate the configuration file

1. Open VSCode

1. For initial VSCode configuration, refer to Teacher Dayu's tutorial Day 2 - the VSCode basic operations section in "How to Modify Locally + AI Interaction" (tutorial link:

• Install the Chinese language plugin. Once you reach the interface shown below, reopen VSCode and it will switch to Chinese

• Select a folder - name it whatever you like. We can create a folder called migpt, until you reach this step shown below:

• If a trust prompt appears, just click "Yes"

1. Create a migpt project terminal in VSCode, entering the state shown on the right

1. Enter git clone followed by your own repository project address copied in step 4:

2. For example: git clone https://github.com/idootop/mi-gpt.git

1. Press Enter to clone the repository project. When you see the screen below, the cloning is complete. Click on mi-gpt in the left sidebar to view the project files

1. Modify the configuration files. The two files with "example" in their names are the ones we need to modify

1. Rename these two files to .env and .migpt.js

Step 6: Modify the vscode configuration file

1. Modify the .env file, refer to the official documentation

1. Here we're using the Qwen 2.5-7b-instruction model, with the following configuration parameters:

1. After modification, it should look like the image below:

1. Fill in the Xiaomi ID and other information we obtained in step 2, corresponding to Xiaomi account ID, login password, and device name, around line 93

1. Modify the MIoT device commands, as shown in the image below

• For my own two devices, the commands I found in the table are:

• I bound the S12 Xiaomi AI speaker, the commands in the table are [5, 1], [5, 3], which match the defaults so no modification is needed. The playingcommand in the table is empty, so no need to add it.

• If it's L15A, then you need to change it to the corresponding code and delete the two // before playingcommand, as shown below

• Remember, all symbols here must use English punctuation, otherwise it will throw an error

1. To modify the persona, you can refer to the official documentation or set it yourself. You can get it running first then modify. Here's a copyable example:

1. Save the file by clicking File in the top-left corner, then click Save All

Step 7: Install dependencies and run

1. In the terminal window below, click Terminal and enter

1. Install dependencies by entering pnpm install

1. Build the project by entering pnpm build

1. To run the project, enter pnpm run dev

2. When running successfully, you'll enter the page shown below. Each conversation will be displayed in the terminal

3. Each time you need to start the service, follow the method shown below to enter the terminal. First enter cd mi-gpt, then run pnpm run dev to start the service

4. Check out the running results. Remember to give the large model time to think - speaking too fast will cause problems