Day = today so far, Pacific time (same window as the Today card above). Week = rolling 7 days.
| Machine | Provider | Calls | Input tokens | Output tokens | Total cost |
|---|
Click a run for the full breakdown. Shows the last 10 days by default - click a workstation's name under Settings to see its complete history instead.
| Machine | Mode | Docs | Auto-resolved | Cost | Started | Duration |
|---|
Read-only organization admin keys, used only to cross-check our own cost tracking against each provider's real billing data once a day. Never used to make inference calls.
For Gemini specifically: this is a Google Cloud API key with the Cloud Billing API enabled on its project - it only reads Google's public pricing catalog (current $/token rates), not your actual usage or spend. Gemini has no usage/billing API equivalent to OpenAI's, so there's no way to reconcile real Gemini spend the way the section below does for OpenAI - this key only powers the "Gemini pricing check" panel further down, which verifies our stored price table against Google's currently published rates.
Per-model $/million-token rates actually used to compute cost. "Default" is what's baked into the broker's code; "Override" lets you correct a rate here without a code change or redeploy when a provider changes pricing - exactly what the reconciliation history below is for catching when it's needed.
| Provider | Model | Code default (in/out) | Currently in effect (in/out) | Edit |
|---|
Runs automatically once daily (06:00 UTC) for the previous UTC calendar day, comparing our internal cost tracking against what the provider actually billed. Deliberately UTC, not Pacific - OpenAI's own Costs API buckets by UTC day, so both sides of the comparison have to use the same window or the numbers won't line up. A UTC day doesn't match a Pacific calendar day (offset by 7-8 hours) - the "Day" column below is UTC, not the server's local time. A small, consistent gap is expected (cached-token discounts aren't tracked separately yet - see code notes); a large or growing gap means something's actually wrong with pricing or tracking.
| Day (UTC) | Provider | Internal total | Real total | Discrepancy | Status |
|---|
Runs automatically once daily (06:00 UTC) AND on every "Run now" below, comparing our stored Gemini pricing against Google's currently published Cloud Billing catalog rate for each model. This is NOT a usage/spend check like OpenAI's above - Gemini has no equivalent billing API - it verifies the price table itself hasn't gone stale. The catalog price is auto-applied as the new override on every run (added 2026-08-22) - no manual "Save" needed in the table above, it updates itself daily and stays current even if a provider changes its rate (e.g. a promotional price ending) without notice. The "Discrepancy" column below shows how much the price actually just moved, not a to-do list item. Requires a Gemini Cloud Billing API key above.
| Checked | Model | Was (in/out) | Catalog / now applied (in/out) | Change | Status |
|---|
Overrides each workstation's own local model settings, pulled automatically at the start of every run. Leave a field blank to let that workstation keep using whatever it has configured locally.
Global, real-time, cross-provider hard stop - the thing no single provider's own spend limit reliably offers.
A Dealer owns both its Locations (a plain label, written into the output index - has no effect on extraction) and its Applications (the actual extraction template - crop regions, regex, validation). Workstations pull this whole tree automatically to build their picker - nothing to touch on any workstation when you change something here.
| Dealer | Locations | Applications |
|---|
| Name |
|---|
| Application | Last updated |
|---|
| Machine | Token | Last connected | Local mode | Status |
|---|