Deploy a DeepSeek AI App with Web Search & Private Knowledge Base in 30 Minutes

PAI+DeepSeek: Build an Intelligent Application with Web Search and Private Knowledge Base

The DeepSeek series of models have gained global attention for their outstanding performance, often matching or surpassing top closed‑source models in inference speed. Since February 2025, Alibaba Cloud AI platform PAI has released best‑practice guides covering deployment, application building, distillation, fine‑tuning, and more, allowing developers to efficiently use DeepSeek‑R1, DeepSeek‑V3, and related models on the cloud.

Step 1: Deploy a DeepSeek‑R1 Model

1. Open the Model Gallery page in the PAI console.

2. After logging in, select a region and workspace, then navigate to Quick Start > Model Gallery ( link ).

3. In the model list, choose DeepSeek‑R1, DeepSeek‑V3, or a distilled version such as DeepSeek‑R1‑Distill‑Qwen‑7B, and submit a deployment task.

Step 2: Use SerpAPI to Enable Web Search

1. Register an account on SerpAPI and obtain an API key.

2. Open the PAI‑LangStudio page in the console.

3. In the left navigation, go to Model Application > Large Model Application Development (LangStudio) .

4. In the "Connection Management" tab, create a new custom connection for SerpAPI.

Step 3: Combine the Model with a Private Knowledge Base

1. Upload enterprise data such as product manuals or FAQs to Alibaba Cloud OSS.

2. In LangStudio, switch to the "Knowledge Base Index" tab, create a new index, and let the system automatically parse the documents to build a dedicated knowledge base, improving answer relevance.

Reference: Knowledge Base Index Management

3. To update the index, click the "Update" action for the target knowledge base.

Step 4: Create an Application Flow in PAI‑LangStudio

1. In the "Connection Management" tab, create a generic LLM model service connection, selecting the model service you deployed in Step 1. The console auto‑fills the base_url and api_key; append /v1 to the base_url.

2. Create a new application flow and select the template "Chatbot with RAG and Web Search".

3. Fill in the flow parameters as shown in the screenshots below.

4. After the runtime starts, click the dialog button in the top‑right corner to begin a conversation; responses will include live web search results.

Step 5: Deploy the Application Flow as an EAS Service

In the LangStudio flow interface, click the Deploy button, specify the service name, resource group, and VPC information, and deploy the flow as an Elastic Algorithm Service (EAS). The resulting service supports both web search and the private knowledge base.

Because EAS services cannot access the public internet by default, configure network connectivity for SerpAPI access following the documentation: Configure Network Connectivity .

For detailed deployment steps, refer to the application flow deployment guide: Application Flow Deployment .

Support and Q&A

Developers are encouraged to join the PAI‑Model Gallery user group via DingTalk (group number 79680024618) for further assistance.