MUIOGO-AI

Run MUIOGO — the UN DESA modelling interface for CLEWs and OG-Core — without the GUI, driven by AI skills: install it, run scenarios, couple it with OG country models, and get analytical outputs on demand.

Works on macOS and Linux. Windows is not supported yet.

Requirements

Git. If you don't have it:

macOS:

xcode-select --install

Linux (Debian/Ubuntu):

sudo apt install git

Install (about 15 minutes)

Open a terminal (macOS: press Cmd+Space, type Terminal, press Enter) and run:

git clone https://github.com/EAPD-DRB/MUIOGO-AI.git

cd MUIOGO-AI

./scripts/install.sh --country PHL

This installs MUIOGO, the Philippines example country, and the OG-CLEWS link, and checks that everything works before finishing.

At the end it offers to install the modelling skills for use outside this repository. You can say no and do it later — see below.

The AI skills

The skills teach an AI assistant how to build, calibrate, run, and review these models. Inside this repository there is nothing to install — open it in Claude Code or Codex (cd MUIOGO-AI, then claude or codex) and the skills are already active. Ask for what you want in plain language, for example "assess the calibration of this CLEWs model" or "run the preflight checks before I start this solve".

New to this? docs/USING_THE_SKILLS.md walks through three worked examples with the exact prompts to type.

To use them in your own model repositories too:

./scripts/install-skills.sh

It asks which assistant you use (Claude Code, Codex, both, or a folder you name) and copies the skills there; restart your assistant afterwards.

See SKILLS.md for the full list and what each one does.

Start MUIOGO

muiogo-ai serve --detach

Leave that window open. For the web interface, open http://127.0.0.1:5102 in your browser. Press Ctrl+C in the terminal to stop.

If something goes wrong

• "port … is already in use" — add --port 5103 to the command. (The installed runtime uses 5102; 5002 is left for a MUIOGO you run yourself.)
• A message about conda — run conda deactivate and try again.
• Still stuck? The installer writes logs into ~/muiogo-ai/ — share the newest .log file when asking for help.

Manual installation

The installer above just runs each project's own installer for you. To do it by hand instead:

1. MUIOGO — follow the install instructions in the MUIOGO README.

2. An OG country model — use the OG-Core universal installer:

   curl -fsSL https://raw.githubusercontent.com/PSLmodels/OG-Core/master/scripts/install.sh -o og-install.sh
   bash og-install.sh --repo og-phl --dest ~/.muiogo/og-models --yes

3. ogclews-link — clone it and run its setup:

   git clone https://github.com/marcelolafleur/ogclews-link.git
   cd ogclews-link && ./scripts/setup.sh --og-path ~/.muiogo/og-models/OG-PHL

For contributors

• Layout: docs/ (scope and design), .agents/skills/ (the skills; .claude/skills/ holds one symlink each), SKILLS.md (the catalogue), client/ (Python client + muiogo CLI), clews/ (country catalog), experiments/ (studies).
• Start with docs/SCOPE.md; install details are in docs/INSTALL_DESIGN.md.
• One hard rule: talk to MUIOGO over HTTP only — never import its backend code. Process is light: push to main, branch when you want review.

License

Apache License 2.0 (LICENSE), same as MUIOGO.