← Back to research
·3 min read·company

OpenSquirrel

OpenSquirrel — native Rust/GPUI control plane for running Claude Code, Codex, Cursor, and OpenCode side by side. Coordinator/worker delegation, remote SSH targeting, persistent sessions. 243 stars, MIT, macOS (Metal).

Key takeaways

  • Native Rust/GPUI control plane for running multiple AI coding agents side by side — Claude Code, Codex, Cursor Agent, and OpenCode in a GPU-rendered tiled layout. No Electron
  • Coordinator/worker delegation — a primary agent (Opus) can spawn sub-agents for focused tasks. Workers return condensed results, not full transcripts
  • Remote machine targeting via SSH + tmux — agents can target local or remote machines. Configure machines in config.toml
  • Persistent sessions — agent state, transcripts, scroll positions, and pending prompts survive app restarts. ~7,200 lines of Rust across 3 files

FAQ

What is OpenSquirrel?

A native macOS app (Rust/GPUI/Metal) for running multiple AI coding agents side by side. Supports Claude Code, Codex, Cursor Agent, and OpenCode with coordinator/worker delegation, SSH remote targeting, MCP integration, and persistent sessions.

How does it compare to Okena?

Both use Rust/GPUI. Okena focuses on terminal multiplexing with cross-platform support. OpenSquirrel focuses on multi-agent orchestration with coordinator/worker delegation and remote SSH targeting. OpenSquirrel is macOS-only (Metal GPU).

Overview

OpenSquirrel is a native, GPU-rendered control plane for AI coding agents built in Rust on GPUI (the UI framework from Zed). It runs Claude Code, Codex, Cursor Agent, and OpenCode side by side in a responsive tiled layout — agents auto-arrange based on count (1=full, 2=split, 4=2x2, etc).

The tagline says it all: "For people who get distracted by agents." Instead of switching between terminal tabs, OpenSquirrel gives you a single view of all your agents working simultaneously.

Key stats: 243 stars (day one), MIT license, Rust, macOS only (Metal GPU). Created March 15, 2026.


Architecture

Multi-Agent Grid

Run multiple agents simultaneously in a GPU-rendered tiled layout. Each agent gets its own pane with structured output parsing — custom markdown rendering with code blocks, diffs, headings, and bullets. All parsed from stream-json output across runtimes.

Coordinator/Worker Delegation

The standout feature: a primary agent (e.g., Opus) can automatically spawn sub-agents for focused tasks. Workers execute and return condensed results, not full transcripts. This enables hierarchical task decomposition without manual coordination — similar to Grok Build's multi-agent approach but across heterogeneous agents (Claude Code + Codex + Cursor simultaneously).

Remote Machine Targeting

Agents can target local or remote machines via SSH + tmux. Configure machines in ~/.opensquirrel/config.toml and assign agents to specific hosts. Run Opus locally coordinating while workers execute on beefy remote servers.

Supported Runtimes

RuntimeModePermission Bypass
Claude CodePersistent stdin (multi-turn)--dangerously-skip-permissions
CodexOne-shot per prompt--dangerously-bypass-approvals-and-sandbox
Cursor AgentOne-shot per prompt--yolo
OpenCodeOne-shot per promptAuto-approved in run mode

Additional Features

  • MCP integration — MCP servers (Playwright, browser-use, etc.) wired directly to agent CLI args
  • Persistent sessions — State, transcripts, scroll positions, and pending prompts survive restarts
  • 8 themes — midnight, charcoal, gruvbox, solarized-dark, light, solarized-light, ops, monokai-pro
  • Command palette — Cmd-K for themes, settings, compact context, kill, views

Technical Details

~7,200 lines of Rust across 3 files:

  • src/main.rs — UI, agent lifecycle, rendering, keybinds, persistence
  • src/lib.rs — Line classification, markdown parsing, diff summarization, helpers
  • tests/state_tests.rs — 30 integration tests

Built on GPUI (standalone crate from Zed), GPU-rendered via Metal on macOS. No Electron, no web views — pure native rendering.


Competitive Position

Strengths: Native Rust/GPUI performance (no Electron). Coordinator/worker delegation across heterogeneous agents. SSH remote targeting. Persistent sessions. MIT license. Clean architecture (~7K LoC).

Weaknesses: macOS only (Metal requirement). Day-old project (March 15, 2026). Requires manual cargo build. Permission bypass flags are security risks. No IDE integration.


Research by Ry Walker Research