Orca: A Multi‑Agent AI‑Native IDE for Parallel Coding

Overview

Orca is an AI‑native Agent Development Environment (ADE) that treats AI agents as first‑class citizens. The platform runs as an Electron 42 desktop application and uses React 19 + TypeScript for the UI, Zustand 5 for reactive state, xterm.js (WebGL) for terminal rendering, Monaco editor for code editing, Vite 7 for builds, Tailwind CSS 4 for styling, and shadcn/ui + Radix UI for accessible components. Git worktree is the basic unit of isolation for each agent.

Design principles

Worktree First – each AI development task runs in its own Git worktree.

Agent‑Native IDE – every IDE capability (terminal, browser, file system, automation) can be invoked by an agent through a standardized interface.

Multi‑Agent Orchestration – the platform supports parallel agents without a single‑window chat limitation.

Core capabilities

Parallel worktrees – each worktree provides isolated file system, terminal, context and session, preventing interference between agents.

Ghostty‑level terminal – WebGL‑rendered xterm.js offers unlimited split panes, persistent state across restarts, full ANSI colour support and ~60 fps performance even with massive output.

Design Mode – a built‑in Chromium instance extracts HTML, CSS and a screenshot of any UI element with a single click and injects that data as context for an AI agent.

SSH worktrees – agents can run on remote servers via SSH with automatic reconnection, port forwarding and secure credential storage.

CLI‑first design – the orca command line exposes almost all IDE functions, enabling scripting and CI/CD integration.

Skills system – developers can add custom orchestration, computer‑use, CLI and Linear‑ticket plugins.

Typical workflows

Parallel solution exploration – define multiple worktrees in orca.yaml (e.g., baseline with Claude, alternative‑1 with Codex, alternative‑2 with OpenCode) and launch them simultaneously with orca orchestrate. Compare results and merge the best implementation.

worktrees:
  - name: baseline
    agent: claude
    task: "Implement core feature"
  - name: alternative-1
    agent: codex
    task: "Try alternative technology"
  - name: alternative-2
    agent: opencode
    task: "Explore third approach"

AI‑assisted code review – create a review worktree, add comments on a diff, then run orca orchestrate review-feedback to let the agent apply fixes or explanations.

# Start review worktree
orca worktree create --agent claude --task "Review PR #2491"

# Add comments on diff
> Potential null‑pointer risk
> Suggest defensive check

# Send feedback to agent
orca orchestrate review-feedback --comments review.md

Remote team collaboration – multiple developers SSH into the same remote server, each operating an isolated worktree while sharing compute resources.

# Developer A
orca ssh user@dev-server --worktree feature/checkout

# Developer B
orca ssh user@dev-server --worktree fix/auth-bug

Technical architecture

Desktop framework: Electron 42

Frontend: React 19 + TypeScript

State management: Zustand 5

Terminal: xterm.js (WebGL) with Ghostty‑level performance

Editor: Monaco (VS Code‑compatible)

Build tool: electron‑vite + Vite 7

UI components: shadcn/ui + Radix UI

Styling: Tailwind CSS 4

Performance and extensibility

Using Git worktree instead of containers provides zero‑overhead isolation, native Git integration, instant switching between worktrees and simple cleanup. The Skills system organizes custom plugins under skills/ (e.g., orchestra/, computer-use/, orca-cli/, linear-tickets/), allowing the platform to adapt to complex development needs.

Key innovations

Worktree First – Git worktree is the default isolation mechanism for AI development.

Agent‑Native IDE – every IDE capability can be called by an agent.

Multi‑Agent orchestration – native support for parallel agents without a single chat window.

Scriptable workflow – the CLI exposes nearly all IDE capabilities for external orchestration.