lessweb/deepcode-cli

Deep Code CLI – a terminal‑based AI coding assistant

What it is – A Node‑JS command‑line tool that lets you chat with DeepSeek LLMs (e.g., deepseek‑flash, deepseek‑v4‑pro) directly from any project folder. It is tuned for DeepSeek’s thinking mode and cache‑aware inference, and can also work with any OpenAI‑compatible model.

Installation

npm install -g @vegamo/deepcode-cli

After the global install, run deepcode in a project directory to start the interactive session.

Configuration – Create ~/.deepcode/settings.json (shared with the companion VS Code extension). Example:

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

The file supports layered overrides (project‑level, user‑level, env vars) – see docs/configuration.md.

Core features

Feature	What it does
Agent Skills	Load custom “skills” (plugins) from ./.deepcode/skills/, ./.agents/skills/, ~/.deepcode/skills/ or ~/.agents/skills/. Skills extend the assistant with extra commands or tools.
DeepSeek‑specific optimisations	Uses DeepSeek’s KV‑cache to cut token cost, enables the model’s thinking mode and lets you control reasoning effort (max, medium, …).
Slash commands & hotkeys	Quick actions like /new (new chat), /resume (pick a past conversation), /model (switch model or thinking settings), /mcp (view connected Model‑Context‑Protocol services), /undo, etc.
Multimodal support	When the selected model can handle images (deepseek‑flash), you can paste an image (Ctrl+V) or give a file path; the CLI will send it as base64 or via DeepSeek Files API.
MCP integration	Connect external services (GitHub, browsers, databases, …) through the Model Context Protocol; manage servers via the mcpServers field and invoke them with /mcp.
Permission control	Fine‑grained policies for shell execution, file I/O, network access – configurable as allow, ask, or deny per‑action type.
Post‑task notifications	Run a user‑provided script (e.g., Slack webhook) after a task finishes, set via the notify field.
Cross‑client sharing	The same settings.json is used by the VS Code extension, so you can switch between terminal and editor seamlessly.

Typical workflow

1. deepcode → start a REPL.
2. Type a prompt (code question, bug description, design request).
3. Use /model to pick a model or toggle thinking mode.
4. Invoke a skill with / menu or a slash command.
5. Accept or reject any generated file or shell actions according to your permission settings.
6. Finish – optional notification script runs automatically.

Extending the tool – Drop JavaScript/TypeScript modules into any of the four skill directories. The CLI scans them in the order shown in the table above, allowing project‑specific or user‑wide extensions without rebuilding the core.

Related projects – A matching VS Code extension (vegamo.deepcode-vscode) shares the same configuration file, and a benchmark repo (deepcode-qrcode-benchmark) demonstrates the performance advantage of Deep Code + DeepSeek on a real‑world Python task.

Development & contribution

git clone https://github.com/lessweb/deepcode-cli.git
cd deepcode-cli
npm install
npm test          # run tests
npm run build      # lint, type‑check, compile
npm link           # link globally for local testing

Before submitting a PR, run npm run check (type‑check + lint) and npm run format.

License – MIT.

---

All details are taken directly from the repository’s README.