Primary Source Commodities

List every commodity instrument in the catalog with sector, unit, primary source, and freshness metadata. Call this first to discover valid codes before price.latest or inventory.get. Free discovery tool — no payment required.

Return the latest cash/fundies observation for one instrument code. Use catalog.list to discover codes such as CL, GC, or ZC. Response follows the shared observation contract with provenance and license tags.

Fetch EIA weekly petroleum inventory levels. Omit series to list all latest inventory points, or pass a series key to get week-over-week change, vs-5y deviation, and a surprise label for that series.

Return the latest CFTC Commitments of Traders managed-money positioning for an instrument code. Includes long, short, net, and week-over-week net change with source URL.

Compute a derived spread observation: 3-2-1 crack, soybean crush, WTI-Brent basis, or calendar (calendar requires licensed curve data and may return available=false). Prefer crack/crush/basis for live derived values with formula and inputs.

Return historical observations for an instrument code (up to 500 points). Use after price.latest when you need a short time series for the same code.

Return the latest FX rate observation for a pair or FRED FX series. Useful alongside commodity prices when converting cash values across currencies.

Fetch a fundamentals document: USDA WASDE, crop progress, export sales, or USGS mineral summaries. Choose kind based on the ags/metals question you are answering.

Build a sector snapshot bundling prices, inventories, spreads, and related fundies for energy, metals, ags, or all sectors. Prefer this over calling many single-observation tools when you need a full sector picture.

List EIA, USDA, and CFTC release calendar entries with cadence, typical Eastern time, and next-hint notes. Free discovery tool — call before scheduling agents around inventory or WASDE prints.

Build a deterministic decision pack for energy, metals, ags, or all sectors. Returns method, signals with direction/rationale, an embedded snapshot, and a disclaimer — no ML model, rule-based only.