Can't use Claude Code, or want to run a terminal coding agent on a Chinese model? DeepSeek's official docs have already given Deep Code a recommended slot. It's an open-source terminal AI coding assistant built for the DeepSeek-V4 model, with support for deep thinking, reasoning-effort control, and Agent Skills. This post covers what it is, what it can do, and how to install it.

What Is Deep Code

The official one-line definition: "Deep Code is an open-source terminal AI coding assistant for the DeepSeek-V4 model."

It's a third-party open-source project (GitHub: lessweb/deepcode-cli, npm package @vegamo/deepcode-cli) built on the Node.js stack, and DeepSeek lists it as one of the recommended options on the Agent integration page of its official API docs. Positioning-wise it mirrors Claude Code — runs in the terminal, can invoke skills, targets the same "write code + modify code" engineering workflow — but the model is DeepSeek-V4, and the config file is shared with the VSCode extension.

⚠️ Note: DeepSeek states explicitly in its docs that this agent is provided entirely by a third party and listed only as a developer reference; the company makes no guarantees about its effectiveness or security. Evaluate it yourself before use.

Core Features

• Deep thinking: on by default for the deepseek-v4 model family, letting the model reason before it acts
• Adjustable reasoning effort: two levels, max or high, controlling how deeply the model "thinks" before acting
• Agent Skills: an extensible skill system supporting both user-level and project-level skills
• Web search tool: can be enabled in the config so the agent can look things up online
• VSCode extension: shares the same ~/.deepcode/settings.json as the terminal CLI — configure once, works in both

Three Steps to Get Started

Step 1: Install

Requires Node.js 18+. Run in your terminal:

npm install -g @vegamo/deepcode-cli

Once installed, verify that it's ready (the official docs mark this step as "verify the installation"; run deepcode --version to confirm).

Step 2: Configure the API Key and Model

Create the config file ~/.deepcode/settings.json:

{
  "env": {
    "MODEL": "deepseek-v4-pro",
    "BASE_URL": "https://api.deepseek.com",
    "API_KEY": "sk-..."
  },
  "thinkingEnabled": true,
  "reasoningEffort": "max"
}

Get an API key from the DeepSeek open platform (platform.deepseek.com). Common model names are deepseek-v4-pro (flagship) or deepseek-v4-flash (lightweight).

Key config options:

💡 Tip: The same config file is reused by the VSCode extension. Set it up in the terminal once and it's directly usable in the IDE.

Step 3: Enter Your Project Directory and Launch

cd /path/to/my-project
deepcode

Handy shortcuts after launch:

How to Use Agent Skills

Skills are Deep Code's extension mechanism, discovered automatically from two locations:

• User-level: ~/.agents/skills/<name>/SKILL.md
• Project-level: ./.deepcode/skills/<name>/SKILL.md

Press / to open the skill picker, or invoke a skill by typing its name (e.g. /skill-writer). That means you can turn your team's recurring workflows (code-style checks, framework-specific scaffolding, documentation generation) into skill files and share them across the team.

Who It's For

• Developers who want to use DeepSeek-V4 for terminal coding but have restricted access to Claude Code
• Engineering teams that want to capture internal workflows as reusable skills
• Users who prefer open-source, self-hostable options and don't want to be locked into a single subscription

The One-Line Takeaway

Deep Code isn't developed by DeepSeek itself, but it fills the gap for the "DeepSeek-V4 + terminal coding agent" combination. Simple to configure, extensible through skills, and sharing settings with VSCode, it's currently one of the most direct ways to run a Claude Code-style workflow on a Chinese model. Just remember it's a third-party project and evaluate the stability yourself.