One IDE for every coding agent. Claude Code, Cursor, Codex, Gemini, OpenCode — same chat, same terminal, same preview, same git. Hot-swap harness mid-session without re-pasting context.
Download the latest macOS build from GitHub Releases.
Signed and notarized by Apple — opens without security warnings.
Building from source works on macOS, Linux, and Windows — see Development below.
Build is a desktop IDE that orchestrates your coding agents. Point it at any project folder and you get:
- Multi-harness orchestration — Claude Code, Cursor Agent, Codex, OpenCode, and Gemini in one window. Switch harness in a click.
- Auto Build — intelligent routing that picks the right harness and model for each stage: plan, build, verify, refine
- Monaco editor — quick search, multi-file tabs, full syntax support
- Integrated terminal — watch every command your agent runs
- Live browser preview — in-window webview with URL bar, DOM inspector, and live reload
- Git UI — branches, diffs, commit history, push/pull
- SSH remote sessions — connect to remote servers and drive agents over SSH
- Session management — multiple projects in parallel, each with their own branch and context
- Semantic search — QMD-powered codebase navigation
- MCP servers — connect Model Context Protocol tools for extended capabilities
- Voice mode — speech-to-speech conversations (optional, requires OpenAI/ElevenLabs keys)
- Download from Releases and drag to Applications
- Open any project folder
- Bring your own API key — Anthropic, OpenAI, or OpenRouter
- Start building
No account required for local-folder mode.
Build doesn't ship its own agent. It's an IDE that drives existing harnesses from one workspace.
| Harness | Provider | Strengths |
|---|---|---|
| Claude Code | Anthropic | Long-horizon refactors, deep tool use, multi-file design |
| Cursor Agent | Cursor | Precision edits, composer-style sessions |
| Codex | OpenAI | Reasoning-heavy tasks, GPT-class and O-series models |
| OpenCode | Community | BYO model — Llama, Qwen, DeepSeek, anything OpenAI-compatible |
| Gemini | Google's models via the Gemini harness |
Hot-swap harness mid-session — the conversation, files, and branch all carry over.
Auto Build orchestrates multi-harness sessions automatically:
- Plan — lead model understands and plans the task
- Build — routes execution to the best-fit harness and model
- Verify — validates the output
- Refine — iterates on feedback
Select "Auto Build" from the model picker, or choose a specific model directly.
| Model | Tier |
|---|---|
| Opus 4.7 | Flagship — most capable |
| Opus 4.6 | Highly capable |
| Sonnet 4.6 | Default — excellent balance of speed and capability |
| Sonnet 4 | Fast and capable |
| Haiku 3.5 | Instant — sub-agents, quick edits |
| GPT-5 / O3 / O4-mini | Via Codex harness |
| BYO local | Via OpenCode — Llama, Qwen, DeepSeek |
Custom models supported via API proxy. Anthropic Foundry (Azure) deployment also supported.
Build never proxies your API calls. Bring your own keys — Anthropic, OpenAI, OpenRouter — and calls go straight to the provider. No telemetry by default.
# Clone and install
git clone https://github.com/Parcha-ai/build.git
cd g-build
npm install
# Run in development mode
./scripts/dev.sh
# Lint
npm run lint
# Build distributable
npm run makeElectron app with a React renderer and Node.js main process:
src/
├── main/ # Main process — services, IPC handlers
│ └── services/ # Auto-router, Claude, Codex, Cursor, SSH, browser, git, etc.
├── renderer/ # React UI — Zustand stores, components
└── shared/ # Types and IPC channel constants
Key technologies: Electron 38, React 18, TypeScript, Zustand, Tailwind CSS, Monaco Editor, xterm.js, node-pty.
MIT — see LICENSE.