Deploy DeepSeek AI: Cloud, Local, API – Full Step‑by‑Step Guide

Deployment Options

DeepSeek can be run in three main ways, each suited to different scenarios:

API (full‑power version) – Quick start without any infrastructure.

Local deployment – Ideal for privacy‑sensitive workloads or when you need full control over the runtime.

Cloud private deployment – Uses Tencent Cloud High‑Performance Computing (HAI) to launch a model in minutes.

API Access

All major cloud providers expose DeepSeek through an OpenAI‑compatible REST API. After creating an API key in the Tencent Cloud console, configure the client as follows:

base_url: https://api.lkeap.cloud.tencent.com/v1

path: /chat/completions

model: DeepSeek-r1

API creation console: https://console.cloud.tencent.com/lkeap

Chatbox client download: https://chatboxai.app/zh

Local Deployment with Ollama

Install Ollama (https://ollama.com) and pull a DeepSeek model that matches your GPU memory. Recommended model sizes:

1.5 B – for machines with only integrated graphics.

8 B – for GPUs with 4 GB VRAM.

8 B or 32 B‑Q4 – for GPUs with 8 GB or more (Apple M1‑M2 Pro/Max can use 32 B).

After installation, start the model with: ollama run deepseek-coder:8b Then use the Chatbox UI (or any OpenAI‑compatible client) to interact with the locally running model.

Cloud Private Deployment (HAI)

Using Tencent Cloud HAI, you can launch a DeepSeek‑R1 instance in about three minutes, avoiding hardware purchases and driver installations. Steps:

Log in to the Tencent Cloud console and create a new HAI application for DeepSeek‑R1.

After creation, a password is sent via internal message.

Choose either the visual ChatbotUI or a JupyterLab console to start interacting with the model.

Cloud Native Build (CNB)

CNB provides a container‑based environment that starts a DeepSeek model in under ten seconds. A free quota of 1 600 core‑hours per month is available (equivalent to a 64 GB machine running for 50 hours).

AI Application Development on Tencent Cloud Knowledge Engine

To build a knowledge‑base Q&A application:

Create a new application in the Knowledge Engine.

Set the model to “DeepSeek‑R1”.

Enable internet search and optionally attach OCR or multimodal services.

Upload internal documents; the platform parses and indexes them automatically.

After configuration, the application can answer queries with citations from the uploaded knowledge base.

Development Tools Integration

Tencent Cloud AI Code Assistant can call DeepSeek via the official API or a locally deployed Ollama instance, providing code completion, real‑time search, and intelligent Q&A inside the IDE.

Enable the assistant:

Install the AI Code Assistant plugin in your IDE.

In the plugin settings, select “DeepSeek” as the model and provide the API key or local endpoint.

Mini‑Program Integration

Three patterns are supported for integrating DeepSeek into Mini‑Programs:

SDK direct model calls – Use the SDK to invoke the model for generic tasks such as text generation or translation.

SDK Agent calls – Invoke an AI agent with custom prompts and knowledge‑base configuration for conversational scenarios.

AI dialogue UI component – Drop a pre‑built UI component into the Mini‑Program with minimal code.

Documentation for Mini‑Program AI capabilities: https://docs.cloudbase.net/ai/miniprogram-using

Example repository: https://github.com/TencentCloudBase/cloudbase-ai-example

Reference Guides

Detailed deployment and usage guides (technical documentation):

DeepSeek deployment on Tencent Cloud: https://cloud.tencent.com/document/product/1721/115966

Knowledge‑base Q&A construction: https://cloud.tencent.com/document/product/1759/116007

Workflow application construction: https://cloud.tencent.com/document/product/1759/116015