Prompt Hospital Brings Theme Hospital Logic to AI Orchestrators

Overview

A new open-source project called Prompt Hospital is bridging the gap between complex AI orchestration logic and intuitive visual management. Drawing inspiration from the classic 1997 simulation game Theme Hospital, the tool provides an isometric control panel for managing teams of AI models. Rather than relying on opaque command-line flags or static configuration files, users can now visually assign tasks, monitor model status, and manage workflows in a way that mirrors physical office dynamics.

The project positions itself as a solution to the growing complexity of AI agent orchestration. As developers increasingly string together multiple language models—such as a specialized coder, a reasoning engine, and a final synthesizer—managing the handoffs between them becomes error-prone. Prompt Hospital attempts to bring clarity to this process by treating each AI model as an employee with a dedicated room and a specific role within a simulated office environment.

How It Works

At its core, Prompt Hospital uses an isometric scene rendered entirely with Vite, React, and TypeScript. The interface is built without external graphics libraries, relying instead on SVG for crisp, offline-capable visuals. Each AI model in a user’s stack is represented by a chibi character sitting in their own office room. These characters display idle animations, such as walking, drinking coffee, or sleeping, depending on their current operational state.

The central interaction mechanic revolves around a task tray, conceptually named after the Roman god Hermes. Tasks appear as envelopes in this coordinator’s tray. Users interact with the system through a simple drag-and-drop interface: dragging an envelope from the tray onto a specific model’s room signals the orchestrator to route that task to the assigned worker. This visual metaphor makes the abstract concept of API calls and background jobs tangible and easier to debug.

Behind the scenes, the system maintains a single source of truth via a file called journal.json. This JSON structure records every model, task, status change, and historical operation. When a user drags a task in the frontend, an HTTP POST request sends the operation to a local or remote server. The server updates the journal, which the dispatcher then reads to execute the actual model calls.

Technical Architecture

The application follows a decoupled architecture that separates the visual interface from the backend logic. The frontend communicates with a lightweight Python API server, typically deployed via Docker and served through nginx with basic authentication for security.

Key components include:

• Frontend (src/): Handles the isometric scene, character states, and drag-and-drop logic. It includes a local store for standalone mode but switches to remote operations when connected to an API.
• API Server (server.py): Serves the current journal state and accepts operations like addTask, assign, unassign, and complete.
• Journal (journal.py): A CLI tool that allows developers to inspect or manipulate the state of the system directly, ensuring that the frontend and backend remain synchronized.
• Dispatcher (dispatch.py): The engine that actually runs the models. It polls the journal for assigned tasks and executes them using backends such as Codex or Claude.

This separation allows the visualization layer to be rebuilt or swapped without affecting the underlying task execution logic. Developers can run the system in a purely local mode for testing, where the journal is stored in localStorage, or deploy it across networks for production-grade orchestration.

Installation and Setup

Getting Prompt Hospital running requires a standard software development workflow. Users begin by cloning the repository and installing dependencies using npm. The frontend development server launches on port 5173, providing a live preview of the isometric office.

For a fully integrated experience, developers must configure the journal URL at build time using the VITE_JOURNAL_URL environment variable. This instructs the frontend to send its operations to the backend API rather than storing data locally.

On the backend side, users copy the example journal configuration to journal.json and start the Python server. The dispatcher can then be launched in a loop to continuously check for new tasks. The project includes detailed documentation in both English and Russian, along with a roadmap that outlines future features such as expanded model support and more sophisticated workflow routing.

Key Insight: The project emphasizes that AI models finally have an "office." This is not just a dashboard; it is a simulation of workflow logic that helps developers understand how tasks flow through their systems in real-time.

Who Is This For?

Prompt Hospital is designed for developers and researchers who are building multi-agent systems. It is particularly useful for those who find traditional debugging methods—such as parsing log files or inspecting raw JSON responses—too cumbersome. By providing a visual representation of task assignment and model status, the tool helps identify bottlenecks, such as a model that is consistently busy or a task that is stuck in the queue.

While currently a proof-of-concept with a small community presence, the project has garnered attention for its creative approach to a common engineering problem. It demonstrates that usability improvements in developer tools often come from borrowing metaphors from unexpected places, such as computer-aided design software or classic simulation games.