An open API service indexing awesome lists of open source software.

https://github.com/akemmanuel/opengui

Desktop + web command center for OpenCode, Claude Code, Codex, and Pi. Multi-project AI coding sessions with streaming chat, prompt queue, model switching, and MCP tools.
https://github.com/akemmanuel/opengui

agent-ui ai-agents claude-code codex coding-agents desktop-app developer-tools electron llm-tools opencode self-hosted

Last synced: 14 days ago
JSON representation

Desktop + web command center for OpenCode, Claude Code, Codex, and Pi. Multi-project AI coding sessions with streaming chat, prompt queue, model switching, and MCP tools.

Awesome Lists containing this project

README

          


OpenGUI


Desktop + web command center for coding agents. Run OpenCode, Claude Code, Codex, and Pi across multiple projects with streaming chat, prompt queue, model switching, voice input, and MCP tools.


Latest Release
License
Stars
Downloads
Build Status


Download latest release
·
Why OpenGUI
·
Supported backends
·
Build from source


OpenGUI Demo

OpenGUI gives coding-agent users desktop and browser workflow for long sessions. Manage multiple projects visually, run different agent backends from one UI, watch responses stream live, queue prompts while agent works, and switch models or agents without terminal juggling.

> Early but usable. Bug reports and PRs welcome.

## Why OpenGUI

OpenGUI is for people who like coding agents but want stronger workflow than terminal tabs alone:

- **Run multiple agent backends in one app** instead of juggling separate tools
- **Manage multiple projects at once** with separate sessions per workspace
- **See streaming responses live** with token and context usage
- **Queue prompts while agent is busy** instead of waiting to type next step
- **Switch providers, models, agents, and variants** from UI
- **Configure MCP tools and skills** without leaving app
- **Use voice input** with Whisper-compatible transcription endpoint

## Highlights

- **Multi-agent workspace** for OpenCode, Claude Code, Codex, and Pi
- **Multi-project workspaces** for parallel coding sessions
- **Real-time streaming** over SSE with live usage tracking
- **Prompt queue** that auto-dispatches when assistant becomes idle
- **Model, backend, and agent selection** directly from chat workflow
- **Slash commands** from prompt box
- **Syntax highlighting** with Shiki
- **Dark/light theme** with system-aware toggle
- **Desktop, web, and Docker deployment options**
- **Cross-platform builds** for Linux, macOS, and Windows

## Supported agent backends

OpenGUI currently supports these coding-agent backends:

- **OpenCode**
- **Claude Code**
- **Codex**
- **Pi**

Use one backend or switch between them per workflow.

## Download

Grab prebuilt app from [latest release](https://github.com/akemmanuel/OpenGUI/releases/latest):

- **Linux:** `.deb`
- **macOS:** `.dmg`
- **Windows:** `.exe` installer

### Requirements

Backend requirements depend on what you use:

- **OpenCode backend:** [OpenCode CLI](https://opencode.ai) installed and available in your `PATH`
- **Other backends:** local CLI/auth/config for that backend available on your machine

> **Windows prerequisite for OpenCode:** OpenCode must be available on your `PATH` or at `%USERPROFILE%\.opencode\bin\opencode.exe`.

> **Note:** Windows builds are unsigned. Windows SmartScreen will warn on first launch. Click **More info** -> **Run anyway**.

## Build from source

### Prerequisites

- [Bun](https://bun.sh) v1.2+ runtime
- [pnpm](https://pnpm.io/) 10+ package manager
- At least one supported backend configured locally (for example OpenCode CLI in your `PATH` for OpenCode)
- [Electron](https://www.electronjs.org/) installed through project dependencies

OpenGUI uses Bun as the runtime for the Electron/web backend, pnpm for dependency management, and Vite+ (`vp`) as the development/build/check task runner.

Install dependencies:

```bash
pnpm install
```

No manual config file needed. Connection settings live in UI. Pick backend, connect workspace, start prompting.

### Development

Run Electron app with HMR:

```bash
vp run dev
```

Run web app with local backend API (projects, git, agents):

```bash
vp run dev:web
```

Open `http://127.0.0.1:3000`. Browser folder picker uses server paths. Set `OPENGUI_ALLOWED_ROOTS=/path/to/projects` to restrict browsable folders.

### Docker

Official image: `ghcr.io/akemmanuel/opengui:latest`.

Docker install supports contained mode and host-control mode. Host-control mode uses host CLIs through `nsenter` while Docker manages web server.

See [docs/docker.md](docs/docker.md) for GHCR install, Docker modes, and [docs/apache.md](docs/apache.md) for Apache reverse proxy + Basic Auth.

### Production

Build frontend bundle:

```bash
vp build
```

Run Electron app in production mode:

```bash
vp run start
```

Build and run web app in production mode:

```bash
vp run start:web
```

For internet-facing deploys, keep OpenGUI bound to localhost and put Apache or another HTTPS reverse proxy in front.

### Distribution

Build Linux `.deb`:

```bash
vp run dist:linux
```

Build macOS `.dmg`:

```bash
vp run dist:mac
```

Build Windows `.exe` installer:

```bash
vp run dist:win
```

## Architecture

```
main.ts Electron main process (window management, IPC)
preload.js Preload script (contextBridge API for renderer)
opencode-bridge.ts IPC bridge to OpenCode SDK
claude-code-bridge.ts IPC bridge to Claude Code SDK
codex-bridge.ts IPC bridge to Codex SDK
pi-bridge.ts IPC bridge to Pi runtime
server/web-server.ts Bun runtime backend for browser mode (RPC, events, server FS browser)
src/
index.html HTML entry point
frontend.tsx React entry point + web Electron shim install
App.tsx Main app layout
hooks/
use-agent-impl-core.tsx Central agent/workspace state
components/ UI components (sidebar, messages, prompt box, etc.)
lib/
web-electron-api.ts Browser shim for Electron preload API
types/ TypeScript type definitions
```

## Configuration

OpenGUI stores connection and UI preferences via the app settings interface.

Voice input (speech-to-text) requires a Whisper-compatible transcription server. Set the endpoint URL in **Settings > General > Voice transcription endpoint**. The microphone button only appears when an endpoint is configured. The server should accept a multipart `POST` with an `audio` file field and return `{ text, language, duration_seconds }`.

## Contributing

Contributions are welcome. See [CONTRIBUTING.md](CONTRIBUTING.md) for guidelines.

## Star History

If you find OpenGUI useful, consider giving it a star -- it helps others discover the project.


Star OpenGUI on GitHub

## License

MIT