MulmoTerminal: Parallel AI Agent Terminal Grid

What changed

MulmoTerminal introduces a novel browser terminal grid specifically engineered for parallel AI coding agent sessions. This tool aims to streamline the workflow for developers interacting with multiple AI agents simultaneously. The grid displays each session in a distinct cell, with a color-coding system to immediately highlight the status of each agent: blue for working, green for done, and amber for sessions requiring user input or attention. This visual feedback mechanism is designed to prevent developers from losing track of which agent is waiting for their next command.

The system supports a range of CLIs, with Claude Code as the default, and also includes first-class support for Codex, Antigravity, Grok, Muse, GitHub Copilot CLI, and Cursor CLI. Each cell functions as a real pseudo-terminal (pty), allowing standard command-line tools like htop, lazygit, or development servers to run within the same environment as the AI agents. This means that a single worktree can host an AI agent alongside other processes, such as a development server launched with yarn dev.

Key features include session persistence via tmux, ensuring that agent sessions survive browser reloads or server restarts. Work is isolated within git worktrees, facilitating easier code management and integration with pull request workflows. Additionally, MulmoTerminal offers a "phone push" notification system that alerts users when a turn finishes or an agent requires input, with a companion "RemoteHost" app allowing for basic interaction directly from a mobile device.

Under the hood, each session runs as a pty on the server and streams to an xterm.js terminal in the browser via WebSocket. The "cockpit roster" provides a real-time overview of all sessions, indicating their state (working, waiting, finished) through visual cues and activity reports. The interface also includes a "Canvas" panel that can render rich output from agents, such as documents, forms, charts, or generated images, moving beyond simple text output.

Why it matters for builders

MulmoTerminal directly tackles the complexity of managing numerous AI agent interactions. By consolidating multiple sessions into a single, visually organized grid, it significantly reduces cognitive load. Developers can now supervise ten agents instead of babysitting one, focusing their attention only where it's needed. The integrated roster and status indicators mean less time is spent figuring out what each agent is doing or waiting for, allowing for quicker decision-making and task progression.

Furthermore, the tool enhances code management by integrating with git worktrees, enabling one-click PRs and diff panels directly from the agent's workspace. This seamless integration between AI agent output and code repository management streamlines the development lifecycle. The ability to receive push notifications on a phone and respond to agent prompts remotely adds a layer of flexibility, allowing developers to stay engaged with their AI assistants even when away from their primary workstation.

Practical impact

Developers can begin using MulmoTerminal immediately by running npx mulmoterminal@latest in their terminal, provided they have Node.js version 22.12 or later installed. The tool defaults to using the claude CLI, but users can configure other agents like Codex via command-line flags or a configuration file. The project is open-source under the MIT license, encouraging community contributions and modifications.

Builders should explore the tool's ability to manage parallel coding tasks, especially when working with multiple AI agents on complex projects. Testing the integration with their preferred CLIs and observing the cost readouts (context %, token counts, estimated $) for each session can provide valuable insights into agent efficiency and resource usage. The git worktree integration offers a practical way to manage code changes generated by different agents working on the same repository.

Caveats and source limits

The provided source material is primarily from the project's GitHub repository, detailing its features and intended use. Specific benchmark results comparing MulmoTerminal's performance against other parallel agent management tools are not available. Pricing information is not mentioned, as it is an open-source project. While the source indicates support for various CLIs, the exact compatibility and setup complexity for each may vary. The "fresh release" status suggests ongoing development, and users should anticipate potential updates and changes to features.