ZCode Beyond

An AI coding workspace — desktop app and zcode terminal CLI — for local LLMs only.

ZCode Beyond is a modified, independent derivative of ZCode, licensed under Apache-2.0. It is not affiliated with or endorsed by the ZCode authors.

Download Install notes LiteLLM setup Updates Source on GitHub

How it differs from ZCode

Local models first: a LiteLLM provider preset (http://localhost:4000/v1) with “Load models” discovery from any OpenAI-compatible server.
No vendor accounts: Z.ai/BigModel login, coding plans and their providers are removed; zcode login/logout are disabled.
No China egress: every process blocks connections to China-operated services (the vendor, Alibaba Cloud, Tencent, ByteDance Lark, Chinese model APIs, any .cn host) before DNS.
No telemetry, no remote control: analytics reporting is off at build time; no auto-update, forced update or remote provider configuration.
Update check on request: Check for Updates… and zcode update ask GitHub whether a newer build exists. Report only — nothing is downloaded or installed.
TUI extras: /review panel for uncommitted changes, /undo / /redo of the last agent turn's file edits, status line with tokens/s, /exit, real context window, and hold-to-delete-faster.

Other providers that are not China-operated (OpenAI, Anthropic, xAI, OpenRouter, custom OpenAI-compatible endpoints) remain available if you add them yourself.

Download

These links always point at the newest build of the main branch (rolling “latest” prerelease, desktop app version 3.14.3). Versioned releases are listed on the releases page.

Builds are unsigned — see install notes. macOS Apple silicon is built for every release; the other platforms are best-effort, so if a link returns “Not Found” that build failed for the latest commit — check the releases page.

macOS your system

Apple silicon (M1 or later) and Intel.

Apple silicon .dmg Intel .dmg

Also as .zip: Apple silicon · Intel

Windows your system

Windows 10/11, x64.

Installer .exe

Linux your system

x64 (glibc 2.28 or newer).

AppImage .deb .rpm Arch

Terminal CLI (`zcode`)

Single executable with the terminal UI and agent — no Node.js needed.

Platform	Download
macOS Apple silicon	zcode-beyond-cli-mac-arm64.tar.gz
macOS Intel	zcode-beyond-cli-mac-x64.tar.gz
Linux x64	zcode-beyond-cli-linux-x64.tar.gz
Linux arm64	zcode-beyond-cli-linux-arm64.tar.gz
Windows x64	zcode-beyond-cli-win-x64.zip
Windows arm64	zcode-beyond-cli-win-arm64.zip
Any (macOS/Linux) with Node.js 24, adds `zcode --web`	zcode-beyond-cli-node.tar.gz

Checksums: every file has a matching .sha256, and all are listed in SHA256SUMS.txt.

Install notes

macOS — unsigned app

Open the DMG and drag ZCode Beyond into Applications. The app is not notarized, so the first launch is blocked: right-click the app → Open → Open. If macOS says the app is damaged, remove the quarantine flag:

xattr -dr com.apple.quarantine "/Applications/ZCode Beyond.app"

Windows — SmartScreen

The installer is unsigned. When SmartScreen shows “Windows protected your PC”, choose More info → Run anyway.

Linux — AppImage, .deb, .rpm

chmod +x ZCode-Beyond-linux-x64.AppImage
./ZCode-Beyond-linux-x64.AppImage

# or Debian/Ubuntu
sudo apt install ./ZCode-Beyond-linux-x64.deb

The packages install as zcode-beyond, separate from an official ZCode install.

Terminal CLI

macOS / Linux (replace mac-arm64 with your platform):

curl -fLO https://github.com/cguldogan/zcode-beyond/releases/download/latest/zcode-beyond-cli-mac-arm64.tar.gz
tar -xzf zcode-beyond-cli-mac-arm64.tar.gz
mkdir -p ~/.local/bin
mv zcode-beyond-cli-mac-arm64/zcode ~/.local/bin/zcode
zcode --help

Make sure ~/.local/bin is on your PATH. If you downloaded the archive with a browser on macOS, clear the quarantine flag once: xattr -d com.apple.quarantine ~/.local/bin/zcode. On Windows, unzip and run zcode.exe.

Node.js build (needs Node.js 24; includes the browser UI via zcode --web):

mkdir -p ~/.local/share/zcode-beyond
tar -xzf zcode-beyond-cli-node.tar.gz -C ~/.local/share/zcode-beyond
node ~/.local/share/zcode-beyond/zcode/bin/zcode.mjs --web

The command is still called zcode and keeps its data in ~/.zcode, like upstream ZCode; check command -v zcode if you also have the official CLI installed.

Verify a download

shasum -a 256 -c ZCode-Beyond-mac-arm64.dmg.sha256   # macOS
sha256sum -c ZCode-Beyond-linux-x64.AppImage.sha256  # Linux

LiteLLM quick setup

LiteLLM is a self-hosted, OpenAI-compatible proxy that can sit in front of your local model servers. ZCode Beyond ships a LiteLLM provider template.

Run a LiteLLM proxy for your local models (its default address is http://localhost:4000).
In ZCode Beyond open Settings → Model settings, add a provider and pick LiteLLM. The base URL defaults to http://localhost:4000/v1; change it if your proxy runs elsewhere.
Enter your LiteLLM key. If the proxy runs without authentication, any placeholder (for example sk-local) works — the field cannot be empty.
Save, then click Load models to import the model names from GET /v1/models, or add model names by hand.

“Load models” works for any provider using the OpenAI chat-completions or responses API, so other local OpenAI-compatible servers can be added the same way.

Staying up to date

Nothing updates by itself. To find out whether a newer build exists, use Check for Updates… in the desktop app (app menu on macOS, Help menu on Windows and Linux) or run:

zcode update

It compares the commit your build was made from with the newest build on this page and lists what changed. Exit code 0 means up to date, 10 an update is available. The check only contacts api.github.com, sends no identifiers and never downloads anything — download and install the new build as above.