esengine/deepseek-reasonix — explained in plain English
Analysis updated 2026-08-13
Run an AI agent in your terminal that edits your codebase only after you approve each change
Keep a long running coding session open so DeepSeek's prefix cache keeps token costs low
Use the desktop app to see which files the agent touched plus live cost and token counters
Continue an active coding session remotely through a connected QQ chat
| esengine/deepseek-reasonix | filebrowser/filebrowser | ehang-io/nps | |
|---|---|---|---|
| Stars | 34,392 | 34,584 | 34,044 |
| Language | Go | Go | Go |
| Setup difficulty | moderate | easy | easy |
| Complexity | 3/5 | 1/5 | 3/5 |
| Audience | developer | general | developer |
Figures from each repo's GitHub metadata at analysis time.
Requires a DeepSeek API key and Node.js 22 or later.
DeepSeek-Reasonix is a command line coding agent built specifically for DeepSeek's models. You install it with npm and run it inside a project folder, where it reads your code, proposes edits, and applies them only when you approve. The tool is designed around a technical detail of DeepSeek's API called prefix caching: if you keep the same conversation running instead of starting fresh sessions, DeepSeek can reuse the earlier part of the prompt and charge much less for it. The README shows a real user's numbers from a single day: 435 million input tokens processed, with 99.82 percent of that traffic served from cache, bringing the bill to around 12 dollars instead of the roughly 61 dollars it would have cost without caching. To use it, you need Node.js version 22 or later and your own DeepSeek API key, which you paste in on first run. The main command is reasonix code, which starts the agent in your project directory. A shorter dsnix alias does the same thing if you prefer fewer keystrokes. Other commands include reasonix chat for plain conversation without file or shell access, reasonix run for one shot tasks that print to the terminal, and reasonix doctor for checking that your setup, meaning Node version, API key, and any connected tools, is working correctly. Beyond the terminal, the project also has a prerelease desktop app built with Tauri that shows the same session in a graphical window, including which files the agent has touched and running cost and token counters. There is also a QQ chat integration that lets you continue a session through the QQ messaging app once connected. This repository is described as the older TypeScript version of the project, now in maintenance mode only for bug fixes. Active development has moved to a separate Go based rewrite on another branch. With over 34,000 stars, it has an active community with a bilingual English and Chinese Discord server for setup help and feature discussion.
A terminal based AI coding agent built for DeepSeek's models, tuned to reuse conversation history so long coding sessions cost much less.
Mainly Go. The stack also includes Node.js, TypeScript, Go.
Setup difficulty is rated moderate, with roughly 30min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Don't trust strangers blindly. Verify against the repo.