Deep Code
Deep Code is a free open-source terminal coding assistant built around DeepSeek V4. Chat in the CLI or VS Code, let it read files, patch bugs, and run commands — billed on the DeepSeek API, not a $20 IDE seat.
The tool is free; tokens are billed on the DeepSeek API. Package names and ~/.deepcode/settings.json follow published Deep Code docs. This is not a $20 IDE seat.
What Deep Code is
Same job as Claude Code: a coding agent in the terminal. Difference: it is open source, tuned for V4 (thinkingEnabled, reasoningEffort), and you pay DeepSeek token rates instead of a closed-model subscription. Surfaces are CLI plus VS Code — not a rented IDE seat.
It is not a new DeepSeek model, and it is not Harness. Harness is DeepSeek’s own agent runtime (npx @deepseek-ai/dsh web). Deep Code is a coding-assistant shell that talks to the DeepSeek API. Install with npm install -g @vegamo/deepcode-cli, then write ~/.deepcode/settings.json. A good default MODEL is deepseek-v4-flash.
When to use Deep Code
Reach for it when you want a free agent shell on V4:
- Edit a repo from the terminal without paying Cursor / Claude Code monthly.
- Turn on V4 thinking and high/max reasoningEffort for harder patches.
- Use the VS Code extension if you want the same agent inside the editor.
- Pay DeepSeek token rates instead of a $20 IDE seat.
What Deep Code is not
Do not put it in the same bucket as models or other agent shells:
- Not a new DeepSeek model. You still call deepseek-v4-flash or deepseek-v4-pro.
- Not Harness. Harness is DeepSeek’s own plugin agent runtime.
- Not Claude Code or Cursor. The tool is free; you are not buying a $20 IDE seat.
- Not the web chat box. Browser chat is for trying prompts first.
- Not the DeepSeek API itself. Deep Code is a shell that calls that API.
At a glance
- What it is
- Free OSS terminal coding assistant
- Surfaces
- CLI + VS Code
- Install
- npm install -g @vegamo/deepcode-cli
- Config
- ~/.deepcode/settings.json
- Default model
- deepseek-v4-flash
- Thinking
- thinkingEnabled + reasoningEffort
- Needs
- Node.js 18+ and a DeepSeek API key
- Billing
- API tokens, not a $20 IDE seat
Three-step setup
Need Node.js 18+. Then:
- Install: npm install -g @vegamo/deepcode-cli
- Write ~/.deepcode/settings.json with your DeepSeek API key, base URL, and a model (deepseek-v4-flash is a good default).
- cd into a project and run deepcode. Shift+Enter for multiline; / opens skills.
{
"API_KEY": "<your DeepSeek API Key>",
"BASE_URL": "https://api.deepseek.com",
"MODEL": "deepseek-v4-flash",
"thinkingEnabled": true,
"reasoningEffort": "high"
}Deep Code vs Harness vs Claude Code vs web vs API
Same API, pick the shell. Open the one you actually need:
| Option | Role | When to pick it |
|---|---|---|
| Deep Code | Free OSS coding assistant, CLI + VS Code, V4-native | You want a terminal / editor agent, not a $20 seat |
| DeepSeek Harness | DeepSeek’s plugin agent runtime (npx @deepseek-ai/dsh web) | You want DeepSeek’s own runtime and plugins |
| Claude Code | Keep Claude Code’s UI; DeepSeek is the brain | You already live in Claude Code |
| Web app | Browser chat | Try copy and reasoning before you wire an agent |
| DeepSeek API | Hosted endpoint for your product | You are shipping, not just chatting |
Boundaries and traps
- The CLI is free; tokens are not. Without a DeepSeek API key in ~/.deepcode/settings.json, Deep Code cannot call a model.
- Need Node.js 18+. Older Node will fail the global install of @vegamo/deepcode-cli.
- Do not treat Deep Code as Harness. Harness is DeepSeek’s own runtime; Deep Code is a coding-assistant shell on the same API.
- This is not a $20 IDE seat. You are not buying Cursor or Claude Code. You pay DeepSeek token rates.
- If published Deep Code docs rename a field in settings.json, those docs win over this page. Do not invent other package names.
Related guides
FAQ
Is Deep Code official DeepSeek software?
It is an open-source coding assistant adapted for V4 and listed in community/API materials. Treat install commands as published by the Deep Code project. The package to install is npm install -g @vegamo/deepcode-cli. It is not a joint Anthropic product, and it is not Harness.
How do I install and configure Deep Code?
Need Node.js 18+. Install with npm install -g @vegamo/deepcode-cli. Write ~/.deepcode/settings.json with API_KEY, BASE_URL (https://api.deepseek.com), MODEL (deepseek-v4-flash is a good default), thinkingEnabled, and reasoningEffort. Then cd into a project and run deepcode. Shift+Enter for multiline; / opens skills.
Flash or Pro inside Deep Code?
Daily edits: deepseek-v4-flash. Hard refactors and long files: deepseek-v4-pro. Change MODEL in ~/.deepcode/settings.json. thinkingEnabled and reasoningEffort (high or max) sit on top of whichever model you picked — they do not replace the model ID.
Do I still need an API key?
Yes. The tool is free; tokens are billed on the DeepSeek API. Get a key from the official platform and put it in ~/.deepcode/settings.json as API_KEY. Deep Code is not a $20 IDE seat and it is not a hosted chat that runs without a key.
Deep Code or Harness?
Deep Code is a free coding-assistant shell: CLI + VS Code, tuned for V4. Harness is DeepSeek’s own agent runtime — files, terminal, plugins, slogan Agent = Model + Harness, start with npx @deepseek-ai/dsh web. Same API, different jobs. Want a terminal / editor coding assistant? This page. Want DeepSeek’s plugin runtime? Open Harness.
Deep Code or Claude Code?
Claude Code keeps Anthropic’s UI and can point the brain at DeepSeek. Deep Code is a V4-native open-source assistant you install yourself. Neither is a $20 IDE seat on this path: you pay DeepSeek token rates. Pick Claude Code if that workflow is already muscle memory; pick Deep Code if you want CLI + VS Code without that subscription.
CLI or VS Code?
Both talk to the same DeepSeek API through the same settings. Use the CLI (deepcode in a project directory) when you want a terminal agent. Use the VS Code extension when you want the same agent inside the editor. Config still lives in ~/.deepcode/settings.json.
What do thinkingEnabled and reasoningEffort do?
They turn on V4 thinking and set how hard the model should work (high or max in published examples). Put them in ~/.deepcode/settings.json next to MODEL. They do not create a new model ID and they do not replace Harness or the DeepSeek API. Flash plus thinking is still Flash; hard reasoning may still want deepseek-v4-pro.