Reasonix

// Overview

Reasonix is an open-source coding agent built around long autonomous runs. One local engine backs four ways in — a terminal CLI/TUI, a desktop app, the browser, and your editor over ACP — so the same session is reachable from wherever you are working. Plan mode, a permission system, a workspace sandbox and per-turn checkpoints are what make leaving it running tolerable: you can read what it did and undo a turn rather than discovering the damage at the end.

Nothing about the model layer is hardcoded. Providers, the agent, the enabled tools and plugins are all declared in a `reasonix.toml`; DeepSeek ships as a preset and any OpenAI-compatible endpoint is a config entry rather than new code. Two models can run together as an executor and a planner in separate, cache-stable sessions. Context maintenance is deliberately cache-aware: startup injects a small stable environment summary, and stale tool output is snipped and pruned before summary compaction, so the prompt prefix stays as byte-stable as possible.

Extensibility runs through two mechanisms — MCP servers contribute tools, prompts and resources, and Extension Protocol v1 sidecars go further, intercepting runtime events, contributing providers and structured UI, and shipping as versioned plugin packages. Distribution is deliberately frictionless: `CGO_ENABLED=0` produces a fully self-contained static binary, cross-compiled to six targets with one command, so the target machine needs nothing installed beyond the binary. Reasonix is MIT-licensed and written in Go.

// What it does

• One local engine, four front ends: CLI/TUI, desktop app, browser and an editor over ACP
• Plan mode, graded permissions, a workspace sandbox and per-turn checkpoints for auditable, undoable long runs
• Config-driven providers — DeepSeek as a preset, any OpenAI-compatible endpoint as a config entry, no hardcoded models
• Optional executor + planner pair running as two separate, cache-stable sessions
• Cache-aware context maintenance: a stable startup environment summary and stale tool output pruned before compaction
• MCP servers plus Extension Protocol v1 sidecars that can intercept runtime events, add providers and ship structured UI
• A single static Go binary, cross-compiled to six targets, with nothing to install on the target machine

// Getting started

The CLI/TUI, the desktop app and the VS Code extension all drive the same local Reasonix engine, so start with the binary.

npm i -g reasonix

# or, on macOS
brew install esengine/reasonix/reasonix

reasonix.toml

reasonix acp

git clone https://github.com/esengine/DeepSeek-Reasonix.git
cd DeepSeek-Reasonix
make build      # -> bin/reasonix
make cross      # -> dist/

Commands and code are distilled from the project's own documentation — always check the official repo for the latest.

// When to use it

• Leave an agent working on a long task and still be able to read every turn and roll one back
• Drive the same agent session from the terminal at your desk and from the desktop app or browser elsewhere
• Run it against DeepSeek's cache-friendly API, or point it at any other OpenAI-compatible endpoint from config
• Extend it past MCP with a sidecar that hooks runtime events and contributes its own UI

// How Reasonix compares

Reasonix alongside other open-source autonomous coding agents tools AI/TLDR tracks, ranked by GitHub stars.