How it works
-
Add models
Each model gets its own room.
-
Drop a task in the tray
The orchestrator hands it out.
-
Watch who works — and who doesn't
Idle models are easy to spot.
What's inside
-
Rooms and models
Claude, GPT, Gemini, Qwen and others, one room each.
-
Task tray
Envelopes queue at the front desk. Drag one to a room to assign it yourself.
-
Journal
Every task, handoff and result, logged.
-
Rooms
Add, remove and rearrange the floor plan.
-
Idling
No task means coffee, a nap or a walk down the hall.
coming
The Warden
The Warden.
Idle for too long? He'll walk you to the front desk himself.
- 14:02 Codex: idle 6 min
- 14:02 Warden → room 3
- 14:03 Codex took T-12
Who it's for
- You run Claude Code and Codex through an orchestrator
- You compare models on the same tasks
- You just want to see what your agents are up to
Connect
git clone https://github.com/s4ps4n/prompt-hospital.gitcd prompt-hospital && npm installnpm run dev # http://localhost:5173
The office reads and writes the journal over HTTP: GET /journal, POST /op. The demo needs nothing.
- Claude
- GPT
- Gemini
- Mistral
- Llama
- Grok
- Qwen
- Codex
- Nemotron
Roadmap
-
done
Office and rooms
Models, tray, journal, idling.
-
done
Own orchestrator
Journal, API, dispatcher.
-
next
Stats
Who worked how much, per model.
-
next
The Warden
Walks the halls, sends idle models to the front desk.