AI usage & cost

ddx PMO uses large language models (LLMs) to classify communications, generate documents, score sentiment, and answer chat questions. Each call is recorded in a ledger with its token counts and cost. The LLM usage view shows that ledger, filtered the way you want.

This page is for organisation admins only. The link is hidden from everyone else, and if a non-admin reaches the page directly it shows a single line saying LLM usage is visible to organisation admins only — nothing else.

Filters

Two controls sit at the top:

Project — All projects, or a single project.
Period — Last 7 days, Last 30 days, Last 90 days, or All time. Defaults to the last 30 days.

The whole view recalculates against the current filters.

What’s shown

Summary cards

Four totals for the current selection:

Total cost (in US dollars)
Calls — how many LLM requests were made
Input tokens — with a breakdown of cache read and cache write tokens underneath (cached prompt content is cheaper to reuse)
Output tokens

By purpose

A table with one row per purpose — the job the call did, such as classifying an email or answering a chat turn. Each row shows calls, input tokens, output tokens, cache read tokens, cache write tokens, cost, and the last activity time.

By model

The same spend split by the model used (for example a Haiku or Sonnet model), with calls, input tokens, output tokens, and cost.

By project

Calls and cost per project. This table only appears when All projects is selected. Two special rows can show up: Tenant-global (no project) for calls not tied to a project, and Deleted project for spend from a project that has since been removed.

Cost per day

A simple bar list of daily cost and call count across the selected period.

The same ledger also calibrates the live cost estimates shown on the project Operations page, where each action card carries a ”$” estimate of what running it will spend.