How to Configure the Web Search Tool in Deep Code

Set "webSearchTool": true in your ~/.deepcode/settings.json file and restart Deep Code to enable real-time web retrieval for up-to-date coding assistance.

Deep Code is a terminal-based AI coding assistant built on the DeepSeek-V4 family of models and distributed through the deepseek-ai/awesome-deepseek-agent repository. It exposes a JSON-based configuration system that controls model behavior, reasoning effort, and optional tools. Enabling the web search capability allows the agent to query the internet and parse fresh results when static training data is insufficient.

Locating the Configuration File

Deep Code reads user preferences from ~/.deepcode/settings.json on every launch. If this file does not exist, you must create it manually.

According to the Deep Code integration guide at line 53 of docs/deepcode.md, this file stores the webSearchTool flag alongside environment variables and runtime options. The path is hard-coded relative to your home directory, so ensure the directory exists before writing the file:

mkdir -p ~/.deepcode

Enabling the Web Search Tool

The web search tool is controlled by the webSearchTool boolean field. When set to true, Deep Code invokes its built-in search module before generating a response, feeding retrieved web data back into the DeepSeek-V4 context window.

Basic Configuration Example

A minimal configuration requires your API credentials and the explicit toggle:

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

Save this to ~/.deepcode/settings.json to activate web retrieval.

Advanced Configuration with Notifications

You can combine web search with notification hooks and alternative model variants. The following example enables web search for the faster deepseek-v4-flash model while triggering a shell script on completion:

{
  "env": {
    "MODEL": "deepseek-v4-flash",
    "BASE_URL": "https://api.deepseek.com",
    "API_KEY": "sk-..."
  },
  "thinkingEnabled": true,
  "reasoningEffort": "high",
  "notify": "/usr/local/bin/deepcode-notify.sh",
  "webSearchTool": true
}

Applying Changes and Verification

Deep Code only reads settings.json at startup. After modifying the file, fully terminate any running Deep Code session and relaunch:

deepcode

To verify the configuration loaded correctly, query the agent for real-time information (e.g., current package versions). If web search is active, Deep Code will indicate that it is retrieving external data before synthesizing an answer.

Summary

• Configuration location: ~/.deepcode/settings.json stores all Deep Code preferences.
• Key field: Set "webSearchTool": true to enable internet retrieval.
• Documentation: The official guide in docs/deepcode.md at line 53 documents this setting.
• Dependencies: Requires valid env credentials (MODEL, BASE_URL, API_KEY) and a restart to take effect.

Frequently Asked Questions

Where is the web search option documented in the source repository?

The webSearchTool field is documented in the Deep Code integration guide at line 53 of docs/deepcode.md. This file outlines all supported runtime flags and their default behaviors.

Do I need to restart Deep Code after editing settings.json?

Yes. Deep Code parses ~/.deepcode/settings.json only during initialization. Changes made while the agent is running are ignored until you restart the executable.

Can I use the web search tool with any DeepSeek model?

Yes, the tool is model-agnostic within the DeepSeek-V4 family. You can pair webSearchTool: true with any MODEL value defined in your env block, such as deepseek-v4-pro or deepseek-v4-flash.

Does enabling web search affect response latency?

Enabling web search increases per-turn latency because Deep Code must query external search APIs, parse results, and feed them into the context window before the LLM generates a response. Adjust reasoningEffort alongside this setting to balance speed and accuracy.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}