GitHub - Backboard-io/Backboard-Desktop: Open-source alternative to Codex, Devin, Claude Code, Higgsfield & Manus. A desktop AI coding agent IDE built on VS Code (Code - OSS): parallel agents, computer use, AI image & video. Use your Claude or ChatGPT subscription or your own API keys. · GitHub
Skip to content

About

Open-source alternative to Codex, Devin, Claude Code, Higgsfield & Manus. A desktop AI coding agent IDE built on VS Code (Code - OSS): parallel agents, computer use, AI image & video. Use your Claude or ChatGPT subscription or your own API keys.

Topics

Resources

Stars

34 stars

Watchers

1 watching

Forks

Latest commit

 

History

4 Commits

Folders and files

Repository files navigation

Backboard Desktop: the open-source AI coding agent IDE

The open-source alternative to Codex, Devin, Claude Code, Higgsfield, and Manus, in one desktop app.

Backboard Desktop is a free, MIT-licensed AI agent workspace for macOS, Windows, and Linux. Run autonomous coding agents, review their diffs, generate images and video, and let agents drive a browser or your computer, all from a single desktop IDE built on Code - OSS (the open-source core of VS Code). Bring your own Claude or ChatGPT subscription, use your own API keys for OpenAI, Anthropic, Google, xAI, and OpenRouter, or run hosted models through Backboard.

License: MIT Platforms Built on Code - OSS GitHub stars

Why Backboard Desktop · Getting started · Features · FAQ · Contributing · License

An open-source alternative to Codex, Devin, Claude Code, Higgsfield, and Manus

Most AI agent tools are closed, tied to one model vendor, and split across separate apps: one for coding, one for autonomous tasks, one for AI video. Backboard Desktop puts them in one open-source IDE you can read, fork, and run yourself.

If you use... Backboard Desktop gives you
OpenAI Codex An open-source coding agent that edits, runs, and tests code in your repository. Use your own ChatGPT subscription or an OpenAI API key.
Claude Code The same agentic coding loop in a full IDE, with a diff review panel. Use your own Claude subscription (Pro or Max) or an Anthropic API key.
Devin An autonomous AI software engineer workflow: many agent sessions in parallel, each on its own branch or git worktree, all in one sidebar.
Manus A general-purpose AI agent that can use a browser and your computer (computer use), search the web, and work through long multi-step tasks.
Higgsfield AI image and video generation in the same chat: create, edit, and iterate on images and video clips next to your code.

Backboard Desktop is an independent project. It is not affiliated with, endorsed by, or sponsored by OpenAI, Cognition, Anthropic, Higgsfield, or Manus; their names are used only to describe comparable features.

Work your way

Switch between two modes from the title bar:

Mode Purpose
Simple A focused chat workspace for coding agents, with session navigation, file changes, and review tools.
Editor The familiar Code - OSS editing experience, including files, terminals, debugging, and extensions.

Features

  • AI coding agents: agents read, write, and run code in your project, with checkpoints you can rewind and a review panel for every change.
  • Bring your own subscription (BYOS): run agents on your Claude Pro/Max or ChatGPT Plus/Pro plan instead of paying twice.
  • Bring your own key (BYOK): OpenAI, Anthropic, Google Gemini, xAI Grok, and OpenRouter keys, stored in your OS keychain.
  • Hosted models through Backboard: Claude, GPT, Gemini, Kimi, DeepSeek, Qwen, GLM, and more from one model picker.
  • Parallel agent sessions: organize work by project, branch, and session; run agents side by side in isolated git worktrees.
  • Computer use and browser use: agents can operate a browser and desktop apps to finish tasks end to end.
  • AI image and video generation: generate and edit images and video from the chat.
  • Skills, MCP, and memory: extend agents with reusable skills and Model Context Protocol servers, and let them remember context across chats.
  • A real editor: the Code - OSS editing experience with files, terminals, debugging, and extensions, in the same window.
  • Shared appearance: use the same theme in Simple and Editor modes.

Agent integrations require the credentials or accounts for the services you choose. Access to hosted services is separate from this repository's source-code license.

Security and credentials

  • Backboard API key: stored in ~/.backboard_ide/config.json. The agent never reads it from disk — the host injects it as BACKBOARD_API_KEY / BACKBOARD_API_URL into the spawned cli-ts environment, and cli-ts checks process.env before touching any file.
  • BYOK (bring your own key) for OpenAI, Anthropic, Google, xAI, and OpenRouter: keys are added on the API Keys page (account menu, or "Backboard: API Keys" in the command palette). The Electron main process is the single owner: it validates a key through the runtime, then saves it as an encrypted secret:// entry in application storage — the same keychain-backed store ISecretStorageService uses — and activates it in the agent host. The host hands the key set to each cli-ts serve process over stdin, never through argv, the environment, or a file.
  • Key-billed models appear in the model picker as their own sections next to the Backboard-routed and subscription models. The selected model is the billing route: adding or removing a key never silently changes what an existing session bills to.
  • Every Backboard credential can be overridden via environment variables for CI or local development. Provider keys deliberately cannot.

Telemetry

  • No analytics keys ship in this repository. Backboard product analytics (PostHog) and ad-conversion reporting stay off unless a build supplies a key: official releases stamp backboardAnalytics.posthogKey into backboard/product.json at build time, and builds from source send nothing. When enabled, analytics follow the telemetry.telemetryLevel setting.
  • Computer use runs the bundled cua-driver. Its usage telemetry key is also supplied at build time (CUA_DRIVER_POSTHOG_KEY), so a source build sends none. In builds that include it, set CUA_DRIVER_RS_TELEMETRY_ENABLED=0, or "telemetry_enabled": false in ~/.cua-driver/config.json, to turn it off.

Getting started

Run from source on macOS

The development launcher below uses the macOS toolchain. Platform packaging instructions are linked under Documentation.

Install these prerequisites:

  • Git to clone the repository.
  • Node.js 24.15.0, pinned in .nvmrc, with npm.
  • Bun to run the TypeScript agent runtime in a development checkout.
  • Xcode Command Line Tools for native dependencies. Install them with xcode-select --install.
git clone https://github.com/Backboard-io/Backboard-Desktop.git
cd Backboard-Desktop/backboard

# If you use nvm:
nvm install
nvm use

# Install the agent runtime's separate dependencies.
(cd cli-ts && bun install)

# Check prerequisites, then build and launch the app.
./scripts/dev.sh doctor
./scripts/dev.sh

If you manage Node with another tool, select the version in .nvmrc before running the commands above and skip the two nvm commands.

The launcher installs editor dependencies when needed, downloads Electron, starts the incremental build, and waits for it to finish before opening the app. The first build can take several minutes. Press Ctrl+C to stop the supervised app and watcher.

Run from source on Windows

Use native Windows PowerShell. Install Git, Node.js 24.15.0 with npm, Bun, and Python 3 on your PATH. Install Visual Studio 2022 Build Tools with the Desktop development with C++ workload, a Windows SDK, and the MSVC, ATL, and MFC Spectre-mitigated libraries for your architecture. Reopen PowerShell after installing the tools.

This fork's install script detects Visual Studio 2022 or 2019; use 2022. The upstream build prerequisites describe the native dependencies, but may target a newer compiler than this checkout.

git clone https://github.com/Backboard-io/Backboard-Desktop.git
Set-Location Backboard-Desktop\backboard

node --version  # should print v24.15.0
python --version
bun --version
npm ci

Push-Location cli-ts
bun install
Pop-Location

# Keep this terminal open while developing.
npm run watch

Wait for the initial client, extension, and Copilot builds to finish without errors. In a second PowerShell terminal, from the repository root:

Set-Location backboard
Remove-Item Env:ELECTRON_RUN_AS_NODE -ErrorAction SilentlyContinue
.\scripts\code.bat

The launcher downloads Electron and prepares built-in extensions. For UI changes, wait for the watcher and run Developer: Reload Window in the app. Restart the app after main-process changes. Stop the watcher with Ctrl+C when finished.

Run from source on Linux

The following native dependency command targets Ubuntu 22.04 x86_64, matching the repository's Linux build image. Other distributions and Ubuntu versions may use different package names. Use a graphical desktop session to launch the app.

sudo apt-get update
sudo apt-get install -y \
  git build-essential pkg-config python3 python-is-python3 \
  libx11-dev libx11-xcb-dev libxkbfile-dev libxkbcommon-dev \
  libxkbcommon-x11-dev libsecret-1-dev libkrb5-dev \
  libgtk-3-0 libnss3 libasound2 libatk-bridge2.0-0 libdrm2 \
  libgbm1 libnotify4 libxdamage1 libxrandr2 libxss1 libu2f-udev \
  dbus-x11 xdg-utils

Install Node.js 24.15.0 with npm and Bun, then clone and build:

git clone https://github.com/Backboard-io/Backboard-Desktop.git
cd Backboard-Desktop/backboard

# If you use nvm; otherwise select the version in .nvmrc yourself.
nvm install
nvm use

npm ci
(cd cli-ts && bun install)

# Keep this terminal open while developing.
npm run watch

Wait for the initial client, extension, and Copilot builds to finish without errors. In a second terminal, from the repository root:

cd backboard
env -u ELECTRON_RUN_AS_NODE ./scripts/code.sh

The launcher downloads Electron and prepares built-in extensions. For UI changes, wait for the watcher and run Developer: Reload Window in the app. Restart the app after main-process changes. Stop the watcher with Ctrl+C when finished.

Connect an agent

Open the app and complete the sign-in or provider setup for your chosen integration. For local Backboard development, you can supply credentials through environment variables in the launch terminal:

macOS:

export BACKBOARD_API_KEY="your-api-key"
./scripts/dev.sh

Linux (with the watcher running):

export BACKBOARD_API_KEY="your-api-key"
env -u ELECTRON_RUN_AS_NODE ./scripts/code.sh

Windows PowerShell (with the watcher running):

$env:BACKBOARD_API_KEY = "your-api-key"
Remove-Item Env:ELECTRON_RUN_AS_NODE -ErrorAction SilentlyContinue
.\scripts\code.bat

The IDE passes resolved Backboard credentials to the agent process. The standalone CLI also supports environment variables; see the CLI runtime guide for its configuration and provider options.

Development

Run these commands from backboard/. The dev.sh supervisor is for macOS; Windows and Linux use npm run watch with the launch commands above.

Command Purpose
./scripts/dev.sh Set up, build, launch, and watch for changes.
./scripts/dev.sh doctor Check the local toolchain.
./scripts/dev.sh setup Install editor dependencies and download Electron.
./scripts/dev.sh logs Follow build output.
./scripts/dev.sh status Inspect the development processes.
./scripts/dev.sh stop Stop the app and watcher.
npm run watch Watch client and extension builds on any platform.
npm run typecheck-client Type-check the main application sources.

After changing renderer code, wait for the watcher to finish and run Developer: Reload Window (Cmd+R on macOS). The macOS supervisor relaunches after main-process changes; restart manually on Windows and Linux. See the development guide for the full workflow and troubleshooting.

To work on the agent runtime separately:

cd backboard/cli-ts  # from the repository root
bun install
bun run dev
bun run validate

Repository layout

Path Contents
backboard/ The Code - OSS fork, application source, built-in extensions, and build tooling.
backboard/src/vs/sessions/ Agent workspace UI, session navigation, and chat surfaces.
backboard/src/vs/platform/agentHost/ Agent process management and provider integrations.
backboard/cli-ts/ The Bun/TypeScript agent runtime, tools, skills, and MCP support.
scripts/ Release and packaging scripts.
docs/ Project documentation, assets, and technical references.

The desktop app runs cli-ts as a child process and exchanges messages over standard input and output. The runtime manages the agent loop and tool execution, while the IDE presents conversations, files, and reviews.

Documentation

Contributing

Bug reports, documentation improvements, and code contributions are welcome.

  1. Search this repository's issues for an existing report. Include your operating system, app version or commit, reproduction steps, and expected behavior when filing a bug.
  2. For substantial changes, open an issue first to discuss the approach.
  3. Fork the repository, create a branch, and keep each pull request focused on one change.
  4. Follow the surrounding code style and layering rules. Run the relevant checks and describe what you tested in your pull request.
  5. Submit your pull request to Backboard Desktop.

Report Backboard-specific issues here. The inherited VS Code contribution guide documents upstream practices and links to Microsoft's project; use those channels for issues or contributions that belong upstream.

Contributors

Backboard Desktop is built by the goats:

Contributions are welcome; see Contributing.

FAQ

Is Backboard Desktop free and open source? Yes. The app is released under the MIT License, and you can build it from source on macOS, Windows, and Linux. Hosted models and subscriptions you connect are billed by their providers.

Is it an open-source alternative to Claude Code and OpenAI Codex? Yes. It runs the same kind of agentic coding loop (read, edit, run, test) in a desktop IDE, and it can use your existing Claude or ChatGPT subscription.

Can it work like Devin or Manus as an autonomous agent? Agents can run many sessions in parallel, each in its own git worktree, and can use a browser and your computer to complete multi-step tasks. You review and approve their changes.

Does it generate images and video like Higgsfield? Yes. Image and video generation runs from the same chat that drives your coding agents.

Which models does it support? Claude (Fable, Opus, Sonnet, Haiku), GPT, Gemini, Grok, Kimi, DeepSeek, Qwen, GLM, and any OpenRouter model, through your subscription, your API keys, or Backboard.

Is it a fork of VS Code? It is built on Code - OSS, the MIT-licensed open-source core of Visual Studio Code, and keeps its editor, terminal, debugger, and extension support.

Credits

Thank you to Microsoft and the Visual Studio Code community. Backboard Desktop builds on their work on the editor, workbench, terminal, debugger, extension infrastructure, and developer tooling. The upstream Code - OSS repository is the foundation of this project.

We also thank the maintainers and contributors of the open-source dependencies used throughout the app. See Third-party notices and individual source files for acknowledgments and license information.

License

Released under the MIT License.

The repository retains Microsoft's upstream copyright and license notice in backboard/LICENSE.txt. Third-party components retain their respective licenses and notices; see ThirdPartyNotices.txt.

About

Open-source alternative to Codex, Devin, Claude Code, Higgsfield & Manus. A desktop AI coding agent IDE built on VS Code (Code - OSS): parallel agents, computer use, AI image & video. Use your Claude or ChatGPT subscription or your own API keys.

Topics

Resources

Stars

34 stars

Watchers

1 watching

Forks

Used by

Contributors

Languages