claude-code-by-agents: Coordinate Coding Agents in Shared Rooms

Overview

claude-code-by-agents, branded Agentrooms, is a native Swift app for coordinating multiple coding-agent instances from a shared room. It addresses the challenge of routing work to agents running on different machines, keeping their activity visible, and sharing context across requests.

The app connects to an OpenAgents workspace, discovers connected agents, and routes messages by name. It is intended for developers or small teams already running agents with @openagents-org/agent-connector; the app is the interface, while the workspace and agent runtimes are separate components.

Key Features

• Coordinate multiple agent instances from one threaded interface.
• Route requests to named agents using @mentions.
• Connect to agents running on local or remote machines.
• Discover connected agents through an OpenAgents workspace.
• Support Claude Code as the first-class runtime, with Codex and custom runtimes described as part of the provider-agnostic model.
• Provide a SwiftUI interface for macOS and iOS, with platform-specific navigation layouts.
• Poll for workspace and thread updates using adaptive intervals rather than SSE or WebSockets.

Use Cases

• A developer can delegate frontend and backend tasks to separate agents and address each by name.
• A solo developer can keep agents on a laptop, Mac mini, or cloud machine accessible from one interface.
• A small engineering team can use a shared workspace to view agent requests and responses in context.
• A developer experimenting with multiple coding-agent runtimes can use the OpenAgents workspace as a common backplane, where the relevant connectors are available.

Getting Started

Download the latest Apple Silicon .dmg from the GitHub Releases page. The app requires a connected OpenAgents workspace and agents set up with @openagents-org/agent-connector. See the README for setup and build instructions.

Alternatives

• Orkas: Orkas uses a local-first Commander LLM to direct specialist agents, rather than coordinating agents through threaded rooms across local and remote machines.
• clowder-ai: Clowder focuses on persistent agent roles, shared memory, and peer review, rather than a native macOS app with threaded workspaces and remote-agent coordination.
• CLI Agent Orchestrator: CAO delegates coding tasks through a local server and isolated tmux sessions, rather than coordinating work through a macOS app and threaded rooms.
• oh-my-claudecode: oh-my-claudecode orchestrates Claude Code workers through skills and terminal sessions, rather than coordinating multiple agents in cross-machine threaded rooms.

Considerations

• This app is a client, not a standalone agent host. It depends on an OpenAgents workspace and separately running agent connectors.
• The documented download is for Apple Silicon, and building from source requires Xcode 16+ and xcodegen.
• The README describes iOS support, but says the TestFlight link is coming with v1.1.
• Version 0.2.x is a full rewrite of the earlier Electron, Deno, and React application. The pre-rewrite stack is preserved on the pre-v1-archive tag.
• The README identifies Claude Code as the first-class path. Codex and custom runtime availability depends on support in agent-connector.
• Updates use adaptive polling, not SSE or WebSockets.