live-inference

name	live-inference
description	Use when starting the HAMLET live-visualization stack — the townlet inference server plus the Vue frontend — to watch trained agents run from checkpoints in a browser.

Live Inference (Visualization)

Two processes, two terminals.

Terminal 1 — inference server

Run from a directory containing checkpoints:

export PYTHONPATH=$(pwd)/src:$PYTHONPATH
python -m townlet.demo.live_inference <checkpoint_dir> 8766 0.2 10000 <config_path>
# Args (positional): checkpoint_dir, port, speed, total_episodes, config_path

The server broadcasts state over WebSocket on localhost:8766.

Terminal 2 — frontend

cd frontend && npm run dev

Open http://localhost:5173. The frontend connects to the WebSocket on mount.

Notes

• Checkpoints must match the config pack's drive_hash — a checkpoint trained under a different drive.yaml will be rejected.
• Rendering mode is substrate-specific (see frontend/CLAUDE.md): spatial grids render in Grid.vue; aspatial universes get the meters-only AspatialView.vue.