What models does Reasonix support?

DeepSeek ships as a built-in preset, but any OpenAI-compatible endpoint can be added as a config entry in reasonix.toml. It also supports running two models together (an executor and a planner) in separate cache-stable sessions.

Do I need the CLI to use the desktop app or VS Code extension?

No for the desktop app, you configure the provider and model directly in the app. Yes for the VS Code extension, which requires the CLI installed first because it starts your local `reasonix acp` backend.

What language is Reasonix written in, and how is it distributed?

It's written in Go and built with CGO_ENABLED=0, producing a single static binary that cross-compiles to macOS, Linux, and Windows on amd64 and arm64.

How do plugins/tools work in Reasonix?

External tools run as subprocesses communicating over stdio JSON-RPC, using an MCP-compatible protocol. Built-in tools self-register at compile time.

Why does prefix-cache stability matter here?

Reasonix is engineered to keep a stable prompt prefix across a session (via a small environment summary injected at startup and pruning of stale tool output), so DeepSeek's prefix caching keeps token costs lower during long-running sessions.

DeepSeek-Reasonix: Terminal AI coding agent

Reasonix is a terminal-based AI coding agent written in Go, distributed as a single static binary. It's built around DeepSeek's prefix cache mechanics, meaning it's designed for developers who run long-lived agent sessions and want to keep token costs down rather than restarting sessions frequently. It targets developers who want a config-driven, model-agnostic agent they can wire into a terminal, a desktop app, or VS Code.

The project ships as a CLI/TUI, a desktop app, and a VS Code extension, all backed by the same local engine. Configuration lives in a reasonix.toml file, so switching providers, models, or tools doesn't require touching code.

Key features

Config-driven setup: providers, agent behavior, tools, and plugins are declared in reasonix.toml with no hardcoded models.
Multi-model composition: DeepSeek ships as a preset, but any OpenAI-compatible endpoint works as a config entry. You can run an executor and a planner model together in separate, cache-stable sessions.
Plugin architecture: external tools run as subprocesses over stdio JSON-RPC (MCP-compatible), while built-in tools self-register at compile time.
Cache-aware context handling: a small stable environment summary is injected at startup, stale tool output gets pruned before compaction, and the built-in tool schema contract is documented for regression review.
Single-binary distribution: CGO_ENABLED=0 builds a static Go binary, cross-compilable to six OS/arch targets with one command. The only external dependency is a TOML parser.
Multiple entry points: use it as a CLI/TUI, a desktop app, or a VS Code extension that talks to the same local engine via ACP.

Ideal use cases

Reasonix fits developers who keep an AI coding agent running for extended sessions and want the underlying model provider's prompt caching to actually reduce cost, rather than resetting context on every call. It also suits teams who want to swap between DeepSeek and other OpenAI-compatible endpoints without rewriting integration code, or who want to run a cheaper executor model alongside a stronger planner model.

It's a good fit if you want a terminal-native workflow with optional desktop or editor integration, and if you're comfortable managing a TOML config file and MCP-style plugin tools.

It's less suited for users who want a fully managed, no-config SaaS chat experience, or who need deep integration with a specific closed-source IDE beyond VS Code/VSCodium/Theia. If your workflow doesn't involve long or repeated agent sessions, the prefix-cache optimization won't matter much and a simpler chat-based tool may be enough.

Installation

Reasonix offers four installation paths depending on how you want to use it.

CLI / TUI, via npm (any OS) or Homebrew (macOS):

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

Prebuilt archives for darwin|linux|windows × amd64|arm64 plus SHA256SUMS are attached to each GitHub release.

Desktop app: download an installer from the official site for macOS (.dmg/.zip), Windows (.exe/.zip), or Linux (.deb/.tar.gz). Windows builds are code-signed via SignPath.

VS Code extension: install the CLI first (Path A), then add the extension from the Visual Studio Marketplace (SivanLiu.reasonix-agent) or Open VSX for VSCodium/Theia. The extension starts your local reasonix acp backend and adds chat, editor context, tool-call approvals, and workspace sessions.

Build from source:

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)

After installing the CLI, get started with:

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

Run /init inside an interactive session to have Reasonix generate project instructions. For the desktop app, download the installer, launch it, and configure a provider and model directly in the UI, no CLI required.