Opensquirrel Agent Control Plane

What This Skill Does

The OpenSquirrel Agent Control Plane is a high-performance, native macOS application built with Rust and the GPUI framework. Unlike Electron-based alternatives, it offers a lightweight and fluid interface for managing multiple AI coding agents concurrently. This control plane allows developers to run Claude Code, Codex, Cursor, and OpenCode side-by-side, effectively creating a unified cockpit for AI-driven software development. It features advanced capabilities like automatic sub-agent delegation, remote SSH machine targeting, persistent session management, and deep integration with MCP (Model Context Protocol) servers, allowing for seamless context sharing across different AI models.

Installation

To get started with OpenSquirrel, ensure you have the Rust toolchain installed on your macOS machine. First, clone the repository from its GitHub source: git clone https://github.com/Infatoshi/OpenSquirrel. Once cloned, navigate into the directory and run cargo build --release. You can launch the binary directly from ./target/release/opensquirrel. For a standard desktop experience, you can create a macOS .app bundle by creating the necessary directory structure in dist/OpenSquirrel.app/Contents/{MacOS,Resources} and copying the binary and icns files accordingly. Note that when running as an app bundle, ensure your environment variables and PATH are correctly configured so that underlying tools like Claude and npx are accessible.

Use Cases

OpenSquirrel is designed for developers who need to orchestrate complex coding tasks. Common use cases include:

1. Parallel Execution: Running a primary coordinator model (e.g., Claude-Opus) alongside specialized worker models (e.g., Claude-Sonnet) to delegate specific sub-tasks like testing or documentation.
2. Cross-Agent Comparison: Evaluating output from Codex and Claude simultaneously on the same codebase to determine the best approach for a refactor.
3. Remote Orchestration: Managing codebases on remote servers via SSH while maintaining a consistent agent session locally.
4. Tool-Rich Environments: Utilizing MCP servers like Playwright for automated E2E testing directly through the agent interface.

Example Prompts

1. "OpenSquirrel, initialize a session using Claude-Opus as the lead and assign the write-tests task to the Claude-Sonnet worker."
2. "Connect to the staging environment via SSH and have the current agent audit the logs for errors while simultaneously drafting a hotfix."
3. "Run a comparison between the current implementation and a new approach using both the Codex and Claude runtimes, then output the diff."

Tips & Limitations

Because OpenSquirrel interacts directly with CLI tools and handles persistent sessions, it is important to configure your ~/.opensquirrel/config.toml carefully. Use the permission_flags settings to limit what agents can execute on your local machine, especially when using 'yolo' or 'dangerously' bypass modes. Keep in mind that as a Metal-rendered application, it performs best on machines with dedicated GPU support. Since the app does not inherit your shell's PATH, if you encounter 'binary not found' errors, launch the application from your terminal to ensure full access to your installed developer toolchain.