Deploying a Private DeepSeek Large Language Model on JD Cloud with Ollama and Knowledge‑Base Tools

DeepSeek has become a hot topic in the AI community, but its official website often suffers from slow responses, prompting many users to seek a private deployment solution.

Deploying a private DeepSeek model offers several advantages: free usage, lower cost, data privacy, no network dependency, flexible customization, performance gains, and no usage limits.

The model comes in two main families: the full‑size "full‑blood" version (e.g., DeepSeek‑R1 with 6710 billion parameters) and several distilled versions ranging from 1.5B to 671B parameters, which are more suitable for local deployment due to lower hardware requirements.

To start, purchase a JD Cloud virtual computer (e.g., 4‑core 8 GB for ¥9.9 × 7 days) and log in to the cloud desktop.

Installing Ollama : download the installer from ollama.com/download , run the installer, and verify the installation by checking the task‑bar icon and executing ollama in a command window.

Configure Ollama to start automatically by opening services.msc , locating the "Ollama" service, setting its startup type to "Automatic", and confirming.

Running DeepSeek models : after Ollama is running, pull and launch a model version, for example:

ollama run deepseek-r1:1.5b – requires ~1.1 GB, suitable for 4 GB RAM machines.

ollama run deepseek-r1:7b – ~4.7 GB, for 16 GB RAM.

ollama run deepseek-r1:14b – ~9 GB, for higher‑end PCs.

… up to ollama run deepseek-r1:671b requiring ~404 GB.

After the command finishes downloading, the model is ready for queries.

Building a local knowledge base with CherryStudio : download CherryStudio, add the DeepSeek‑R1 model via an API key, upload documents (e.g., the novel "Three‑Body"), and enable the knowledge base in the chat interface for context‑aware answers.

Enabling network search with Page Assist : install the Page Assist browser extension, select the DeepSeek model, enable RAG settings, and toggle the network switch to allow the local model to perform web searches.

Using AnythingLLM : install AnythingLLM, configure it to use Ollama as the LLM backend, set up LanceDB for vector storage, upload documents, and adjust retrieval settings; it also supports optional network access for fallback queries.

These steps collectively provide a complete workflow for privately deploying, running, and extending DeepSeek models on a personal cloud computer, offering both offline capability and optional internet connectivity.