de-agent

A terminal-based AI Coding Agent inspired by Claude Code. Built with TypeScript, designed to assist you in your local development environment.

Features

• ReAct Agent Loop: Automatically reasons, calls tools, observes results, and iterates until the task is complete.
• 7 Built-in Tools: Read/write/edit files, run shell commands, grep regex, glob search, and list directories.
• Multi-Model Support: Compatible with OpenAI-style APIs, allowing you to use Claude, DeepSeek, GPT-4o, or local Ollama models.
• Secure Execution: Safe read-only commands run automatically, while dangerous operations (like rm -rf) require explicit user confirmation.
• Streaming Output: Real-time markdown rendering and tool execution UI directly in your terminal.
• Dynamic System Prompt: Automatically injects Git status, current working directory, OS info, and project memory (.de-agent.md).

Installation

You can install de-agent globally via npm:

npm install -g de-agent

Or run it directly using npx:

npx de-agent

Setup

Before using de-agent, you need to provide an API key. By default, it expects an OpenAI-compatible API.

export DE_AGENT_API_KEY="your-api-key"

Custom Model & Base URL (Optional)

If you want to use a different model or provider (e.g., DeepSeek, Claude via proxy, or Ollama):

export DE_AGENT_MODEL_NAME="deepseek-chat"
export DE_AGENT_BASE_URL="https://api.deepseek.com/v1"

You can also create a config file at ~/.de-agent/config.json:

{
  "apiKey": "your-api-key",
  "baseUrl": "https://api.deepseek.com/v1",
  "modelName": "deepseek-chat",
  "permissionMode": "ask"
}

Usage

Interactive REPL Mode

Launch the interactive prompt to chat and collaborate with the agent:

de-agent

One-Shot Task Mode

Pass a prompt directly to execute a specific task and exit:

de-agent "Create a new Express.js hello world server in the current directory"

CLI Options

• --api-key <key>: Override the API key.
• --base-url <url>: Override the API base URL.
• --model, -m <model>: Specify the model name to use.
• --auto-approve: Automatically approve all tool calls (use with caution!).
• --help: Show help information.

Project Memory

If you create a .de-agent.md file in the root of your project, the agent will automatically read it and include its contents in its system prompt. This is useful for providing project-specific instructions, coding guidelines, or architectural context.

License

MIT