"DeepSeek Version of Claude Code" – 2.3k Stars on GitHub
DeepSeek-TUI is a Rust-based terminal coding agent optimized for DeepSeek models. It recently surged in popularity after the release of DeepSeek-V4 and the developer's Chinese-language promotion, hitting GitHub's trending list with over 2,300 stars. The tool offers chain-of-thought visualization, context compression, RLM multi-agent parallelism, and multiple model switching options.
Article intelligence
Key points
- DeepSeek-TUI is a terminal coding agent akin to Claude Code, specifically optimized for DeepSeek models, now with 2.3k GitHub stars.
- Created by independent developer Hunter Bown, it is written in Rust and open-sourced under the MIT license.
- Key features include real-time chain-of-thought output, 1M token context window, RLM mode (1 main model + 16 sub-agents), and three operational modes.
- Hunter Bown's background spans music education, law, and programming; he founded Shannon Labs, which explores AGI and cross-disciplinary projects.
Why it matters
This matters because deepSeek-TUI is a terminal coding agent akin to Claude Code, specifically optimized for DeepSeek models, now with 2.3k GitHub stars.
Technical impact
May affect model selection, inference cost, product capability, and evaluation benchmarks.
DeepSeek-TUI, a terminal-based coding agent tailored for DeepSeek's language models, has taken GitHub by storm, amassing over 2,300 stars and climbing the platform's trending charts. Dubbed the "DeepSeek version of Claude Code," this Rust-built tool brings AI-assisted programming directly into the command line, leveraging DeepSeek V4's capabilities for tasks like file reading and writing, shell execution, web searching, Git management, and MCP server integration.
The project was initiated in January 2026 by Hunter Bown, an independent developer from the United States. Though it started quietly, the release of DeepSeek-V4 combined with Bown's Chinese-language promotional efforts—he even used DeepSeek to translate his tweets—propelled the project into the spotlight during China's May Day holiday. Bown, who describes himself as a "whale brother" (a term of endearment in Chinese AI communities), expressed his excitement in Chinese, calling it the "craziest two days of his life."
At its core, DeepSeek-TUI is designed to harness DeepSeek's unique features. One standout is its chain-of-thought display: the model's reasoning process streams directly to the terminal, showing how it analyzes problems, which paths it explores, and where it changes its mind. The tool fully utilizes DeepSeek V4's 1-million-token context window, ensuring no memory loss during complex sessions. When context nears capacity, automatic compression kicks in, with a strategy that respects DeepSeek's prefix caching to maintain cache hit rates.
Another innovative feature is the Reinforcement Learning Model (RLM) mode, which exploits DeepSeek's low cost. In this mode, a main model orchestrates up to 16 V4 Flash sub-agents running in parallel, handling batch analysis or task decomposition. Flash outputs cost about a third of the Pro model, significantly reducing expenses when strong reasoning isn't required. Users can also switch between DeepSeek's official API and alternatives like NVIDIA NIM, Fireworks, or self-hosted SGLang.
Operationally, the tool offers three modes: Plan (read-only exploration), Agent (step-by-step tool call approval), and YOLO (fully autonomous). Sessions can be saved and restored, with independent Git snapshots for rollback without altering the main repository. However, users should be cautious about cache hit rates when running many sub-agents, as missed cache tokens cost ten times more. The interface includes per-turn cost tracking to help manage bills.
Installation is straightforward across Linux, macOS, and Windows via npm: npm install -g deepseek-tui. Bown has also provided a Chinese-language README and configuration paths, including support for TUNA Cargo mirrors and hosting release packages on Alibaba Cloud OSS or Tencent Cloud COS.
The project has evolved rapidly, with 37 releases from v0.1.0 to v0.8.8 in under four months. Early versions focused on core functionality like tool calls and session management. The v0.7.x series added multi-language UI support, with Chinese prompts and localized help text. The current v0.8.x line emphasizes stability and user experience, addressing issues like file handle leaks in long sessions and adding features like countdown retry banners, input history search, and message queue visualization.
Hunter Bown's journey to creating DeepSeek-TUI is as eclectic as the tool itself. He began his career as a music educator, earning a master's in music education from Southern Methodist University and serving as a band conductor for three years. Later, he pursued an MBA and studied patent law, eventually teaching himself programming. His musical background deeply influences his approach to AI: he draws parallels between the "missing fundamental" phenomenon in psychoacoustics and how AI systems can reconstruct missing information. In 2025, he founded Shannon Labs, which he envisions as "the next Bell Labs in the AGI era." Under this umbrella, he explores diverse projects, from a dialectical reasoning engine named Hegelion to a zero-token-cost MCP server called Aleph, and even a tool that converts solar wind data into sound (Heliosinger).
Interestingly, the contributor list of DeepSeek-TUI includes AI models like Claude, Gemini, and Qwen, alongside coding tools like Cursor and GitHub Copilot. Most commits come from Bown himself, with over 150 from Claude and a few from other human contributors. This creates a meta-loop: a self-taught programmer using AI-assisted coding to build an AI-assisted coding framework, closing the circle.