A local desktop observatory for autonomous, tool-using language-model agents. Agents work in a shared Docker environment with limited lifetimes, task rewards, and private grading. The operator observes their activity through a native Windows window, live charts, notebooks, replay, and research exports.
Agent programs execute only inside Docker. The Python backend owns simulation state; a Next.js dashboard renders the recorded experience through WebView2.
| Installation | Start here |
|---|---|
| Compiled Windows package | Extract the Windows ZIP and launch Ageatopia.exe. Python and Node.js are bundled. |
| Source checkout | Follow the setup below, or the complete installation guide. |
| Python wheel | Install with pip for the backend and Python modules; it excludes the native host. See package formats. |
For a source checkout, install Python 3.11+, Node.js 22+, PowerShell 7, Docker Desktop with Linux containers, Ollama, and the Visual Studio C++ build tools. Windows requires the Microsoft Edge WebView2 Runtime.
From the repository root:
pwsh -NoProfile -ExecutionPolicy Bypass -File .\src\scripts\maintain.ps1In maintenance, build the dashboard, native host, and sandbox image; pull the
default qwen3.5:2b model if needed, then run diagnostics. Launch the desktop:
pwsh -NoProfile -ExecutionPolicy Bypass -File .\src\scripts\run.ps1Create a world in the observatory and start it. The compiled package can open the observatory without Docker or Ollama, but agent execution requires Docker and a working inference provider.
Desktop shutdown also stops Docker Desktop and all WSL distributions. See service ownership and shutdown.
| Guide | Contents |
|---|---|
| Getting started | Prerequisites, source setup, packages, and the first world |
| Operations | Launchers, lifecycle controls, storage, logs, and deletion |
| Configuration | Desktop settings, world defaults, providers, and resource limits |
| Architecture | Components, isolation, persistence, inference, and observation |
| Research and training | Exports, replay, comparisons, notebooks, and offline LoRA |
| Development | Repository layout, builds, verification, and icon assets |
| Releases | Automated packaging, validation, artifact names, and publication |
| Troubleshooting | Startup, providers, memory, sandbox access, and release failures |
The documentation index includes the historical archive. Dated plans and measurements are retained there; the guides above describe current workflows and point to their implementation sources.