Orca Introduces TypeScript-Based Desktop ADE to Coordinate Fleets of Parallel AI Coding Agents

Stably AI has released Orca, an open-source, MIT-licensed desktop environment built in TypeScript to orchestrate and run fleets of parallel AI coding agents. The graphical user interface (GUI) provides a unified control plane for command-line interface (CLI) agents, replacing scattered terminal windows with integrated workspace management. By isolating active agents within dedicated Git worktrees, the platform aims to eliminate local directory conflicts, file lockouts, and overlapping Git branch modifications.

Isolating Agent Workspaces with Automated Git Worktrees

Command-line coding agents run autonomously to write code, execute tests, and modify files within developer directories. As developers increasingly run models like Claude Code, Codex, or Cursor CLI side-by-side to automate engineering tasks, orchestrating their concurrent operations introduces technical hurdles. If multiple agents attempt to modify the same repository simultaneously, it can trigger file lockouts, redundant edits, and overlapping branch conflicts.

Traditionally, developers must manually stash their code, juggle branches, and monitor several terminal windows. This manual overhead slows down development and complicates the verification of agent-produced solutions.

To resolve these state collisions, Orca implements automated Git worktrees. A Git worktree allows developers to attach multiple independent working directories to a single repository. Instead of switching branches back and forth within a single folder—which forces developers to save unfinished changes continually—worktrees construct isolated directories on the local drive.

When a developer issues a prompt in Orca, the environment can fan the instructions out to multiple AI agents simultaneously. To prevent agents from overwriting each other’s active work, the platform dynamically assigns each agent to an isolated Git worktree.

The output from these sandboxed directories flows directly into a single consolidated dashboard. Developers can monitor live terminal outputs side-by-side, inspect code changes through a built-in diff viewer, and merge the selected implementation once verified.

Developers can also manage these environments programmatically. The platform includes a command-line interface called Orca CLI that supports dedicated commands for orchestrating worktrees, such as:

orca worktree create --repo id:<repoId> --name my-task --issue 123 --json

Mitigating Hardware Load with Agent Hibernation and Remote SSH Support

Running multiple command-line agents in parallel places a significant computational burden on local hardware. Each active agent session operates as a pseudoterminal process that consumes local system memory and CPU threads.

To protect host computers from performance degradation, Orca features an “Agent Hibernation” mechanism designed for resumable coding-agent sessions. When a worktree tab becomes inactive because the user switches focus, Orca kills the idle background agent process to free up local RAM and CPU. When the developer returns to the tab, the platform restarts the process and resumes the session using the agent’s native resume command, such as the native resume command in Claude Code.

For intensive operations that exceed local hardware capabilities, such as compiling large codebases, the platform supports secure shell (SSH) worktrees. This architecture allows developers to run resource-heavy processes on high-performance remote servers while keeping the control interface on their local desktop. The IDE systematically creates the Git worktree on the remote server’s file system, routing input and output streams back through a secure tunnel.

Architectural Separation: Node.js Daemon and React-Vite UI

Under the hood, Orca is built as a cross-platform desktop application using an Electron foundation. The backend infrastructure runs a Node 24 runtime engine, while the visual interface is powered by React and Vite.

This architecture splits execution workloads to prevent user interface lag during heavy agent tasks. A main background process, or daemon, handles CPU-intensive jobs like file system monitoring, remote procedure calls (RPC), and local data transactions.

By separating file system operations from visual rendering, the React-Vite thread remains free to handle visual updates. This division of labor ensures that the user interface remains responsive and minimizes input lag, even when multiple background agents are actively generating code.

Built-In Utilities for Prompting, Visual Review, and Mobile Monitoring

The platform incorporates several tools aimed at streamlining visual tasks, code review, and resource tracking:

• Monaco-Powered Editor: The built-in text editor is built using the Monaco Editor—the browser-based editor component that powers VS Code. It features universal autosave and allows developers to drag images or files directly into prompt inputs.
• Design Mode: An embedded Chromium browser includes a “Design Mode” feature to assist agents with front-end tasks. Developers can select any visual element on an active web page to automatically capture its HTML, CSS, and a cropped screenshot, sending those assets directly to the agent’s prompt context.
• Inline Diff Comments: For code reviews, developers can leave comments on specific lines of an AI-generated diff. The platform sends these annotations back to the active agent, allowing it to modify the code based on human instructions without requiring the developer to leave the application.
• Account and Usage Tracking: An interface tool monitors API token usage, tracks rate-limit resets for networks like Claude and Codex, and supports hot-swapping different user accounts without logging out.
• Mobile Companion App: The desktop application links with a mobile companion app available for iOS and Android. This mobile interface alerts developers when a background agent completes a task, allowing them to review progress and issue follow-up commands on the go.

Scriptable Workflows and WebGL Terminal Splits

For terminal-centric workflows, Orca includes Ghostty-class split terminals. The terminal windows render using WebGL technology, enabling high-performance processing of logs and text while supporting infinite horizontal and vertical splits. The terminal’s scrollback history also survives system restarts.

To enable programmatic control, the platform features the dedicated Orca CLI utility. This tool allows developers to automate multi-agent setups using shell scripts. With commands such as orca worktree create, orca terminal list, orca terminal split, and orca terminal wait, developers can build automated pipelines that provision isolated environments and drive agent behavior without manual interaction.

The CLI also supports integrations with external software development platforms. It pulls issue logs and task assignments directly from GitHub and Linear, allowing engineers to open isolated worktrees directly from a ticket and push the resolved branches back to their continuous integration (CI) pipelines.