DeepSeek-Reasonix: Reasonix – a DeepSeek‑native AI coding agent for your terminal, desktop, or VS Code

DeepSeek‑Reasonix – an AI‑powered coding assistant for the terminal

What it is – Reasonix is a single‑binary Go program that talks to any OpenAI‑compatible LLM (with DeepSeek as the default) and turns the model into a coding agent you can run from the command line, a desktop UI, or inside VS Code. It is driven entirely by a reasonix.toml configuration file and optional plugins, so you can swap providers, add tools, or chain two models (planner + executor) without recompiling.

Key ideas from the README

  • Config‑driven – providers, models, enabled tools and plugins are declared in a TOML file; no hard‑coded model IDs.
  • Multi‑model & composable – you can point Reasonix at any OpenAI‑compatible endpoint and even run two models in parallel (e.g., a planner model that creates a plan and an executor model that writes code).
  • Plugin system – external tools communicate via stdio JSON‑RPC (MCP‑compatible). Built‑in tools are compiled in, but you can add your own.
  • Cache‑aware context – Reasonix injects a stable environment summary at startup and prunes stale tool output, keeping token usage low by re‑using DeepSeek’s prefix cache.
  • Zero‑friction distribution – a statically linked Go binary (CGO_ENABLED=0) that can be cross‑compiled to six target platforms (macOS/Windows/Linux × amd64/arm64). The only runtime dependency is a TOML parser.

Installation options (as described in the README)

Path How to get it
CLI / TUI npm i -g reasonix (pulls the pre‑built binary for any OS) or brew install esengine/reasonix/reasonix on macOS. Pre‑built archives for all platforms are also attached to each GitHub release.
Desktop app Download the installer from the official site https://reasonix.io/?download=desktop#start. Packages are provided as .dmg/.zip (macOS), .exe/.zip (Windows, code‑signed), and .deb/.tar.gz (Linux).
VS Code extension Install SivanLiu.reasonix-agent from the VS Marketplace (or Open VSX for VSCodium/Theia). The extension launches the local reasonix acp backend; you must have the CLI installed first.
Build from source ```bash
git clone https://github.com/esengine/DeepSeek-Reasonix.git
cd DeepSeek-Reasonix
make build # → bin/reasonix(.exe)
make cross # → dist/ (all platforms) 

---

### Quick start (CLI/TUI)
```bash
# 1. Configure your LLM provider and model
easonix setup

# 2. Open an interactive session
reasonix

# 3. Ask it to do work, e.g. implement TODOs in a file
reasonix run "implement the TODOs in main.go"

Inside the interactive session you can type /init to let Reasonix generate a project‑level instruction set.

Main features (summarised from the README)

  • Config‑driven architecture – everything (provider, model, tools, plugins) lives in reasonix.toml.
  • Composable multi‑model pipelines – run a planner and an executor model side‑by‑side.
  • Plugin‑driven tool execution – external tools communicate via JSON‑RPC over stdio; built‑in tools are self‑registered at compile time.
  • Cache‑aware context handling – stable environment summary, pruning of stale tool output, and compacted session summaries to keep token usage low.
  • Single static binary – no CGO, easy cross‑compilation, only a TOML parser required at runtime.
  • Multiple front‑ends – same engine powers a CLI/TUI, a native desktop app, and a VS Code extension.

Documentation & community

  • Guides & references – docs/GUIDE.md, docs/CLI.md, docs/CONFIG_PATHS.md, plus deeper docs on sub‑agents, session memory, capability diagnostics, checkpoints, etc.
  • Community – bilingual Discord server (#help / #求助) for setup help and workflow sharing.
  • Support – optional donations via PayPal or WeChat Pay (no impact on issue triage).

Who’s behind it

The project is MIT‑licensed and maintained by the esengine community. The README lists the top 20 contributors and shows a contributor‑rock‑image, indicating an active open‑source team.

Bottom line – Reasonix is a genuine, production‑ready AI coding agent that lets developers harness DeepSeek (or any compatible LLM) directly from their terminal or IDE, with a focus on low token cost, extensibility via plugins, and easy distribution as a single binary.

Sources