DeepSeek Engineer 🐋

Overview

This repository contains a powerful coding assistant application that integrates with the DeepSeek API to process user conversations and generate structured JSON responses. Through an intuitive command-line interface, it can read local file contents, create new files, and apply diff edits to existing files in real time.

Key Features

1. DeepSeek Client Configuration

   • Automatically configures an API client to use the DeepSeek service with a valid DEEPSEEK_API_KEY.
   • Connects to the DeepSeek endpoint specified in the environment variable to stream GPT-like completions.

2. Data Models

   • Leverages Pydantic for type-safe handling of file operations, including: • FileToCreate – describes files to be created or updated.
     • FileToEdit – describes specific snippet replacements in an existing file.
     • AssistantResponse – structures chat responses and potential file operations.

3. System Prompt

   • A comprehensive system prompt (system_PROMPT) guides conversation, ensuring all replies strictly adhere to JSON output with optional file creations or edits.

4. Helper Functions

   • read_local_file: Reads a target filesystem path and returns its content as a string.
   • create_file: Creates or overwrites a file with provided content.
   • show_diff_table: Presents proposed file changes in a rich, multi-line table.
   • apply_diff_edit: Applies snippet-level modifications to existing files.

5. "/add" Command

   • Users can type "/add path/to/file" to quickly read a file's content and insert it into the conversation as a system message.
   • Users can also type "/add path/to/folder" to add all files in a directory (excluding binaries and hidden files).
   • This allows the assistant to reference the file contents for further discussion, code generation, or diff proposals.

6. Conversation Flow

   • Maintains a conversation_history list to track messages between user and assistant.
   • Streams the assistant's replies via the DeepSeek API, parsing them as JSON to preserve both the textual response and the instructions for file modifications.

7. Interactive Session

   • Run the script (for example: "python3 main.py") to start an interactive loop at your terminal.
   • Enter your requests or code questions. Enter "/add path/to/file" to add file contents to the conversation.
   • When the assistant suggests new or edited files, you can confirm changes directly in your local environment.
   • Type "exit" or "quit" to end the session.

Getting Started

1. Prepare a .env file with your DeepSeek API key: DEEPSEEK_API_KEY=your_api_key_here

2. Install dependencies and run (choose one method):

   Using pip

   pip install -r requirements.txt
   python3 main.py

   Using uv (faster alternative)

   uv venv
   
   uv run main.py

3. Enjoy multi-line streaming responses, file read-ins with "/add path/to/file", and precise file edits when approved.

Reasoning-Enabled Version (r1.py)

A new script r1.py has been added that uses DeepSeek's Reasoning Model (deepseek-reasoner). This version includes Chain of Thought (CoT) reasoning capabilities:

• Shows the AI's reasoning process before providing the final answer
• Maintains all original features (file operations, diff editing, etc.)
• Displays thought process in a dedicated panel
• Only uses final conclusions in conversation history
• Run it with python3 r1.py or uv run r1.py for an enhanced experience with visible reasoning

Note: This is an experimental project developed by Skirano to test the new DeepSeek v3 API capabilities. It was developed as a rapid prototype and should be used accordingly.

Model Selector and API Configuration

The application now supports selecting different models and configuring their respective APIs, including DeepSeek, OpenAI, local OpenAI compatible endpoints, and Mistral.

How to Use the Model Selector

1. When you run the script (python3 main.py), you will be prompted to select a model:

   • 1. DeepSeek
   • 2. OpenAI
   • 3. Local OpenAI Compatible Endpoint
   • 4. Mistral

2. Enter the number corresponding to your choice.

3. The application will configure the client with the appropriate API key and base URL based on your selection.

Configuring Different APIs

1. DeepSeek

   • Ensure you have the DEEPSEEK_API_KEY set in your .env file.
   • The base URL for DeepSeek is pre-configured as https://api.deepseek.com.

2. OpenAI

   • Ensure you have the OPENAI_API_KEY set in your .env file.
   • The base URL for OpenAI is pre-configured as https://api.openai.com.

3. Local OpenAI Compatible Endpoint

   • Ensure you have the LOCAL_OPENAI_API_KEY and LOCAL_OPENAI_BASE_URL set in your .env file.
   • The base URL should point to your local OpenAI compatible endpoint.

4. Mistral

   • Ensure you have the MISTRAL_API_KEY set in your .env file.
   • The base URL for Mistral is pre-configured as https://api.mistral.com.

By following these steps, you can easily switch between different APIs and leverage the capabilities of each within the same application.