Getting Started

Web UI login

1. Log in according to the instructions in Step 9 of the "Quick Deployment", using the initial username admin and the initial password (master_key) set in the "Quick Deployment".

   Figure 1 Logging into the LiteLLM Management Web UI

   

   

   If you forget the password, you can go to the /opt/litellm directory on the server and check the master_key configured in the config.yaml file.

2. Click "Models+Endpoints" on the left to enter the model configuration interface. There are some system-provided example configurations that can be saved or deleted as needed.

   Figure 2 Model Configuration Interface

   

Integration with MaaS services

1. Log in to the MaaS platform. This document takes deploying the DeepSeek-v3 in China-Hong Kong as an example. In the left navigation bar, select "Online Inference".
   Figure 1 MaaS platform

   

2. In the preset service list, select the desired service and click "Activate Service" in the operation column on the right. When grayed out, it indicates that the service has been activated.
   Figure 2 Activate Service
   

3. After successful activation, select the desired service in the service list, click "Usage Instructions", and choose "Standard Interface" as the interface type in the pop-up window to obtain the API address and model name.
   Figure 3 Usage Instructions
   

4. When invoking model services deployed on MaaS, you need to fill in the API Key for interface authentication. In the left navigation bar, click "API Key Management" (you can create up to 30 keys. Each key is only displayed once upon creation, so please ensure it is properly saved. If a key is lost, it cannot be retrieved, and you will need to create a new API Key to obtain a new access key).
   Figure 4 API Key Management
   

5. On the "API Key Management" page, click "Create API Key" in the upper-right corner, fill in the label (custom label for the API Key; labels must be unique and cannot be duplicated. Only uppercase and lowercase English letters, numbers, underscores, and hyphens are supported, with a length range of 1–100 characters) and description (custom description for the API Key), then click "OK." The label and description information cannot be modified after creation.
   Figure 5 Create API Key
   

6. Click "Add Model" to configure the following parameter

   Figure 3 Add Model

   

   Table 1 Basic Configuration Description

   Parameters	Description	Example
   provider	Model Provider (e.g., OpenAI, Bedrock, etc.)	Custom OpenAI
   LiteLLM Model Name	Model Name sent to the API by LiteLLM	deepseek-v4-flash
   Model Mappings	Model Mapping	deepseek-v4-flash
   Mode	Interface Mode	Chat - /chat/completions
   API Base	URL of the API Address	https://api.modelarts-maas.com/v2
   API Key	API Key	step 5 API Key obtained

7. Click "Test Connect". When the content shown in the figure below appears, you can click "Add Model" to add the model.

   Figure 4 Test Connect successful

   

Create a Virtual Key for model management and authentication.

The owner of the Virtual Key can be yourself, a specific account, or an individual user.

1. Click "Virtual Key" and select "Create New Key".
   Figure 6 Create Virtual Key
   
2. Fill in the required information according to the table below, then click OK and copy the Virtual Key information.

   Table 2 Key related information

   Module	Parameter Name	Options/Example	Meaning Description
   Key Ownership	Owned By	You / Service Account / Another User	Key Owner Entity - determines who owns the key: You: Current logged-in user (individual developer scenarios) Service Account: For CI/CD pipelines or automated scripts Another User: Assign to existing LiteLLM team members
   	Team	Search or select a team	used for multi-tenant/team scenarios. The key inherits the team's budget, model permissions, and other configurations, enabling cost and permission management at the team level.
   Key Details	Key Name	deepseek-v4-flash	Custom Alias for the Key - used only for administrative identification, making it easier to distinguish keys for different purposes (e.g., project name, application name). It does not affect API calls.
   	Models	deepseek-v4-flash	List of Models Accessible by the Key - multiple selections allowed. Restricts the key to calling only specified models, preventing unauthorized use of model resources.
   	Key Type	LLM API	Key Type - defines the category of API endpoints it can access: LLM API: General large language model interfaces (e.g., chat/completions, completions) Other optional values typically include specialized API types such as embedding, image, audio, etc., for granular control.

3. In the "Playground", select "Virtual Key" in the "Virtual Key Source" field, then enter the Virtual Key and other related information. In the dialog box, type "hello" to test if it can respond to your message correctly.
   Figure 7 enter the Virtual Key and other related information
   

The Virtual Key is generated only once, please make sure to save it promptly.

For more advanced usage tutorials, Refer to the LiteLLM official website and LiteLLM code repository.