DeepSeek Engineer

DeepSeek Engineer is an open-source Python project by GitHub user Doriandarko that brings AI-assisted coding directly to the terminal. It connects to DeepSeek's API and uses the DeepSeek-R1 reasoning model to provide an interactive pair-programming experience through natural conversation and automatic function calling. The project is licensed under the MIT License and has accumulated over 2,200 stars on GitHub since its creation in December 2024.

What It Is

DeepSeek Engineer is a command-line coding assistant that wraps DeepSeek's reasoning API in a rich terminal interface. Unlike simple chatbot wrappers, it exposes a set of file-operation tools — read, create, edit — that the AI can invoke automatically during a conversation. The result is a workflow where a developer can describe a task in plain language and watch the model reason through it, call the right file tools, and deliver working code or edits without manual copy-pasting.

Function Calling Architecture (v2)

Version 2.0 replaced the original structured JSON output approach with native function calling. The README describes this as a significant upgrade that enables:

• Natural conversations without rigid response formats
• Automatic file operations triggered by the AI when it detects a need
• Chain of Thought (CoT) visibility — the model's reasoning stream is shown in real time before it acts
• Triple-stream processing — reasoning, content, and tool calls are handled simultaneously during streaming

The available tools the AI can call include read_file, read_multiple_files, create_file, create_multiple_files, and edit_file (snippet-based precise replacement).

Terminal Interface and Workflow

The interface is color-coded (green for success, red for errors, yellow for warnings) and shows structured diff previews and progress indicators. A typical session flows as:

1. User types a natural language request
2. The model's reasoning process appears in real time (CoT)
3. The AI executes one or more function calls automatically
4. Results are fed back into the conversation
5. The AI delivers a follow-up response based on the operation results

Users can also manually preload files or entire directories into the conversation context using the /add command, though the README notes this is optional since the AI reads files automatically when they are mentioned.

Security and Safety Design

The project includes several built-in safeguards:

• Path normalization and directory traversal protection
• File size limits (5 MB per file)
• Binary file detection and exclusion
• Automatic directory creation with safety checks

Setup Path

DeepSeek Engineer requires Python 3.11+ and a DeepSeek API key from the DeepSeek Platform. Installation is straightforward via uv (recommended) or pip. A .env file holds the API key. The project uses a standard requirements.txt and supports uv sync for dependency management.

Update: Version 2.0 and Current Status

The repository was created on December 26, 2024, and last pushed to in May 2025, with the metadata record updated as recently as August 2026. Version 2.0 is the current release, marked by the shift to native function calling. The project is described by its author as experimental — built to explore DeepSeek's reasoning model capabilities — and contributions are welcomed via the standard GitHub fork-and-PR workflow.