Auspex is a small prediction harness for supply chain operations. Connect a dataset, choose a backend, get predictions with confidence intervals attached - no pipeline, no feature engineering, no training run.
It handles two task shapes: tabular prediction for delay risk, quality and supplier scoring, and time-series forecasting for demand, inventory and freight. Zero-shot foundation models sit behind the same interface as classic methods, so moving from exploration to production scoring is a config change rather than a rewrite.
The harness exposes itself as an MCP server, which means any agent can call prediction as a tool. Apache-2.0 licensed.