omnirush-ai/omnirush-gui

omnirush‑gui – a desktop coding‑assistant

What it is

• A native desktop application (macOS Apple Silicon and Linux) that lets you chat with a “coding agent” – an LLM that can write, edit, and run code on your local machine.
• The default model is GPT 6 Astra (a proprietary “frontier” model). You can switch to any model in your omnirush.ai account, including GPT 6.1 Sol, or add your own OpenAI/Anthropic keys.

Key capabilities (as described in the README)

Feature	What it does
Model & effort selection	Pick a model and an “effort” level (low → max) per chat, controlling how much compute the agent may use.
Skills & MCP servers	Load reusable skills (pre‑written tool wrappers) and MCP (model‑call‑protocol) servers from a built‑in library; every model sees the same set.
Git workflows	The agent can clone repos, create branches, worktrees, commit, push, open PRs, and even review them via the gh CLI. Destructive actions are confirmed, and the agent asks only once per session for routine commands.
Approvals toggle	A global switch that moves the agent from “ask before every tool call” to “full‑permissions” mode, useful for power users.
Self‑updating	On macOS and Linux the app can download and install newer releases from within the UI.
Cross‑platform packaging	Distributed as DMG (macOS), .deb/.rpm/.pacman packages, AppImage, and tar.gz for Linux, each with a SHA‑256 checksum.

How to get it

1. Visit the Releases page on GitHub or the download buttons on the README to fetch the appropriate package for your OS.
2. macOS: open the DMG (you may need to clear the quarantine flag with xattr -dr com.apple.quarantine …).
3. Linux: install the package for your distro or make the AppImage executable (chmod +x). Install GTK runtime libraries if you use the AppImage/tarball.
4. On first launch, sign in via the Sign in to omnirush.ai button – a browser window opens to omnirush.ai/console where you approve a device code.

Running the app

• After sign‑in, the sidebar lets you start a new chat, pick a model, and adjust effort.
• Use the Library in Settings to add skills or connect to MCP servers.
• The UI will prompt for confirmation before any potentially destructive tool call, unless you enable the full‑permissions toggle.

Development & contribution

• Built with Node.js 24, pnpm 11.4 and Bun 1.3.10+. The source tree contains a monorepo; the desktop client lives in @omnirush/desktop.
• Typical workflow:

  pnpm install --frozen-lockfile
  pnpm --filter @omnirush/desktop dev   # launch the app locally
  pnpm typecheck                         # run TypeScript checks
  pnpm --filter @omnirush/desktop test  # run the desktop test suite
  

• Release builds are automated via a GitHub workflow triggered on v* tags.

Support & licensing

• Bugs/feature requests: open an issue on the GitHub repo.
• General questions: email info@omnirush.ai.
• Security disclosures: email with subject “security”.
• The core desktop code is MIT‑licensed; an ee/ directory contains enterprise‑only code under a separate license.

Bottom line Omnirush‑gui is a genuine AI‑powered desktop tool that brings a high‑performance coding LLM into a native UI, with built‑in Git automation, extensible skills, and cross‑platform distribution. It is aimed at developers who want an on‑device assistant without leaving their local development environment.