DeepSeek-Reasonix is an MIT-licensed, DeepSeek-native AI coding agent shipped as a single static Go binary whose one local engine can be driven from a terminal, a desktop app, a browser, or an editor over ACP — built for developers who want long autonomous coding runs they can still read and undo.

What it is

DeepSeek-Reasonix is an open-source AI coding agent written in Go and licensed under MIT. It compiles with CGO_ENABLED=0 into a fully self-contained static binary, and one command cross-compiles it to six targets covering darwin|linux|windows across amd64|arm64. A single local engine sits at the centre, and there are four ways into it: the CLI/TUI, an official desktop app, a browser, and an editor over ACP. The project is config-driven: providers, the agent itself, enabled tools, and plugins are all declared in reasonix.toml. DeepSeek ships as a preset rather than a hardwired dependency, and any OpenAI-compatible endpoint is a config entry rather than new code. It lives in the MCP and ACP ecosystem, installed through npm and Homebrew, with a companion VS Code extension published to the Visual Studio Marketplace and the Open VSX Registry.

The concrete problem it addresses is session durability rather than raw model quality. The project is engineered around prefix-cache stability and describes itself as something you leave running. It solves that with cache-aware context maintenance: startup injects a small stable environment summary, stale tool output is snipped and pruned before summary compaction, and the built-in tool schema contract is documented so regressions can be reviewed. Long autonomous runs stay governable through plan mode, permissions, a workspace sandbox, and per-turn checkpoints. The thing it replaces is the hardcoded model assumption — the provider, agent, tool set, and plugins are declared in reasonix.toml, so swapping or composing models does not require patching the agent.

Key capabilities

• Config-driven setup: providers, the agent, enabled tools, and plugins all declared in reasonix.toml, with no hardcoded models.
• Multi-model and composable: DeepSeek as a preset, any OpenAI-compatible endpoint as a config entry, and an optional two-model arrangement with an executor and a planner in separate, cache-stable sessions.
• Plugin-driven tooling: MCP servers contribute tools, prompts, and resources, while Extension Protocol v1 sidecars intercept runtime events, contribute Providers and structured UI, and ship versioned plugin packages.
• Cache-aware context maintenance: a small stable environment summary at startup, stale tool output snipped or pruned before summary compaction, and a documented tool schema contract for regression review.
• Long-run controls: plan mode, permissions, a workspace sandbox, and per-turn checkpoints that keep an autonomous run readable and undoable.
• Four front ends on one engine: CLI/TUI, desktop app, browser, and editor over ACP, where the VS Code extension starts the local reasonix acp backend.
• Zero-friction distribution: prebuilt archives for darwin|linux|windows × amd64|arm64 with SHA256SUMS on every GitHub release.

Who uses it and how

• Terminal-first developers install with npm i -g reasonix on any supported platform, or brew install esengine/reasonix/reasonix on macOS, then leave the agent running in a long session.
• Editor-centric developers add the SivanLiu.reasonix-agent extension from the Visual Studio Marketplace or Open VSX Registry, after completing the CLI install, because the extension does not bundle the CLI.
• Teams wanting a two-model workflow run the executor and planner together in separate cache-stable sessions.
• Plugin and integration authors build MCP servers or Extension Protocol v1 sidecars that contribute Providers, structured UI, and versioned plugin packages.
• Desktop users take a .dmg or .zip on macOS, an installer .exe or portable .zip on Windows, or a .deb or .tar.gz on Linux; Windows installers are code-signed through SignPath.io.

Getting started

Install the native binary through npm with npm i -g reasonix on any supported platform, or use Homebrew on macOS with brew install esengine/reasonix/reasonix; prebuilt archives and SHA256SUMS ship on every GitHub release, and the desktop app is available from the official download page.

How it compares

The facts provided list no paid products that this project replaces, and they name no comparable coding agents either. On the record given here, DeepSeek-Reasonix stands alone in this registry.

When to use it — and when not to

A self-hoster runs a local engine and maintains a reasonix.toml, and building from source requires Go 1.26+ with GOTOOLCHAIN=auto kept in place, while the desktop build adds Node 24+ and pnpm 10. Anyone who wants a hosted service with no local binary or config file to manage should look elsewhere, and the VS Code extension requires the CLI to be installed first. The one caution visible in the facts is the backlog: 1,908 open issues against 2,399 forks is a heavy ratio, though the repository was last pushed in September 2026 and the README excerpt is truncated at the quick-start command list, so the full CLI surface is not documented here.

npm i -g reasonix                  # any OS; pulls the prebuilt native binary
brew install esengine/reasonix/reasonix   # macOS

git clone https://github.com/esengine/DeepSeek-Reasonix.git
cd DeepSeek-Reasonix

make build      # -> bin/reasonix(.exe)
make cross      # -> dist/ (darwin|linux|windows × amd64|arm64)

scripts/desktop-build.sh darwin/arm64 v0.0.0-dev   # one platform per run

reasonix setup                      # configure a provider and model
reasonix                            # start an interactive session
reasonix run "implement the TODOs in main.go"