Training runs in the standalone Python repo charmcast-ml (configure the path via the CHARMCAST_ML_PATH env var). Each run writes a status JSON that this tab reads.

Sanity check on synthetic data:

.venv\Scripts\python.exe scripts\train_vol_forecaster.py --synthetic --symbol SPY

On real data (parquet of {ts, price, size}):

.venv\Scripts\python.exe scripts\train_vol_forecaster.py ^
    --ticks data\raw\spy-2026-04.parquet ^
    --bar-kind volume --threshold 1500000 --symbol SPY

See docs/getting_started.md in the charmcast-ml repo for the full guide.

Use these for off-origin integrations like the CharmCast for TradingView Chrome extension. The token replaces a cookie session and authenticates every API call. You'll see the token only once on creation — copy it before closing the modal. Revoke any time.

The CharmCast desktop app (Electron) handles options-chain grabs and forwards them to this server through a reverse-tunnel. It identifies itself with a relay secret stored locally in its .env. Issue one here, paste it into the desktop's .env as RELAY_SECRET=, restart the desktop app — that's it.

RELAY_SECRET=...