# DeepSeek-Reasonix

> Open-source coding agent shipped as a single Go binary with a TUI, desktop app, and ACP editor integration; DeepSeek preset, any OpenAI-compatible model.

Stars: 36.0k · Language: Go · License: MIT · Categories: AI

- Repository: https://github.com/esengine/DeepSeek-Reasonix
- Website: http://reasonix.io/
- Page: https://www.trmn.sh/tools/deepseek-reasonix

## Install

- Homebrew: `brew install esengine/reasonix/reasonix`
- npm: `npm i -g reasonix`
- Build from source: `make build`

## Overview

Reasonix is an open-source (MIT) coding agent for software engineering tasks, built as a single Go binary. One local engine is exposed through four front ends: a terminal CLI/TUI, the Reasonix Studio desktop app, a browser, and editors over the Agent Client Protocol (ACP). A VS Code extension starts the local `reasonix acp` backend and adds chat, editor context, tool-call approvals, model selection, and workspace sessions.

Behavior is driven by a `reasonix.toml` file that declares providers, the agent, enabled tools, and plugins, with no hardcoded models. DeepSeek is included as a preset, and any OpenAI-compatible endpoint can be added as a config entry. Two models can optionally run together as an executor and a planner, in separate sessions that keep their prompt caches stable. Plan mode, permissions, a workspace sandbox, and per-turn checkpoints are meant to keep long autonomous runs reviewable and reversible.

Extensibility comes from MCP servers, which contribute tools, prompts, and resources, and from Extension Protocol v1 sidecars. Sidecars can intercept runtime events, contribute providers and structured UI, and ship as versioned plugin packages. Context maintenance is cache-aware: startup injects a small stable environment summary, and stale tool output is snipped or pruned before summary compaction. The binary builds with `CGO_ENABLED=0` as a static executable, and `make cross` produces darwin, linux, and windows builds for amd64 and arm64.

Typical CLI usage is `reasonix setup` to configure a provider and model, `reasonix` for an interactive session, and `reasonix run "<task>"` for a one-shot task. The `/init` command creates project instructions. The project has two lines: 2.x (Studio, in pre-release) and 1.x (stable, maintenance). It is aimed at developers who want a configurable agent they can run locally against DeepSeek or other OpenAI-compatible models.

### Installation

```bash
# via Homebrew
brew install esengine/reasonix/reasonix
```

```bash
# via npm
npm i -g reasonix
```

```bash
# via Build from source
make build
```

## Alternatives

- [Crush](https://www.trmn.sh/tools/crush.md): Terminal-based agentic coding assistant from Charm that connects to many LLM providers, uses LSPs for context, and extends via MCP servers. (Go, 28.9k stars)
- [Claude Squad](https://www.trmn.sh/tools/claude-squad.md): Terminal UI that runs multiple AI coding agents (Claude Code, Codex, Gemini, Aider) in parallel, each in its own tmux session and git worktree. (Go, 8.7k stars)
- [kubectl-ai](https://www.trmn.sh/tools/kubectl-ai.md): Terminal assistant that turns natural-language requests into kubectl and bash operations using Gemini, OpenAI, Bedrock, Ollama and other LLM providers. (Go, 7.6k stars)
- [mods](https://www.trmn.sh/tools/mods.md): CLI tool that adds AI capabilities to command line pipelines by processing stdin through LLMs (Go, 4.5k stars)
- [Ollama](https://www.trmn.sh/tools/ollama.md): CLI tool for running and managing local large language models with REST API and client library support (Go, 185.1k stars)

---

Canonical: https://www.trmn.sh/tools/deepseek-reasonix