esengine/DeepSeek-Reasonix

Reasonix – Open‑source AI coding assistant

What it is – Reasonix (formerly DeepSeek‑Reasonix) is a locally‑run AI‑powered coding agent. You point it at a project folder, describe a task in plain language, and it reads the code, edits files, runs commands or tests, and shows you each step. The actual language model is not bundled; you configure any OpenAI‑compatible endpoint (DeepSeek preset is provided) and pay the model provider directly.

How you use it

• Desktop app (Reasonix Studio) – a graphical UI that guides you through setup, model selection and task execution.
• Terminal / TUI – the reasonix CLI offers an interactive full‑screen session or one‑off commands (reasonix run "task").
• Web UI – reasonix web starts a local web interface.
• Editor integration – a VS Code extension (and any editor that supports the Agent Client Protocol) connects to a local reasonix acp backend, letting you chat with the assistant right inside the editor.

All four front‑ends share the same underlying engine.

Key features

• Read‑edit‑run loop – the agent can inspect your code, modify files and execute shell commands or test suites.
• Permission levels – you can require the assistant to ask before each action, or let it run autonomously within limits you set (plan mode, forbid rules).
• Rewind / checkpoints – before any file change the original is saved, so you can roll back by conversation turn without touching git history.
• Config‑driven – providers, tools, plugins and policies are declared in a reasonix.toml file; no hard‑coded models.
• Multi‑model support – run a single model or compose two (e.g., a planner and an executor) in separate cached sessions.
• Plugin architecture – MCP servers and Extension‑Protocol sidecars can contribute tools, prompts, UI components, or whole skill packages.
• Static single‑binary distribution – the Go engine builds to a self‑contained binary for macOS, Windows and Linux (amd64/arm64).

Installation

• Reasonix 2.x (Studio) – download the appropriate installer (.dmg, .exe, .deb, .rpm) from the Studio releases page. The package includes the CLI binary as well.
• Reasonix 1.x (stable) – install the CLI via npm (npm i -g reasonix) or Homebrew on macOS, and get the desktop app from the official download page.
• From source – clone the repo, switch to the studio branch for 2.x, then make build (CLI) or make studio (requires Node 24+ and pnpm 10) to produce the binaries.

Typical workflow

1. Install the app or CLI and run reasonix setup to add a model provider and API key.
2. Open a project folder (or launch Studio and select it).
3. Describe the desired change, e.g. "add retry on failure to the HTTP client".
4. The assistant proposes edits and commands; you approve or reject each step. Use the rewind feature if something goes wrong.

Who it’s for

• Developers who want an AI pair‑programmer that works offline on their own machine.
• Teams that need a configurable, self‑hosted coding assistant without vendor lock‑in.
• Contributors interested in extending the platform via plugins or custom tools.

License & community

• MIT license – free for commercial and personal use.
• Active community on Discord, QQ, and GitHub Discussions.
• Sponsorship options (Right Code, PackyCode) provide routing to commercial LLMs.

---

Reasonix lets you keep the code and the model under your control while still benefiting from AI‑driven code generation and automation.