Cost, usage & linked work
Understand estimated AI cost, work allocation, and coverage.
Centrail connects captured AI usage, estimated cost, and associated work. Start with where usage went, then inspect the evidence behind a number.
Estimated AI cost
Reported tokens are priced using model rates, with separate rates for input, output, and supported cache classes. The Overview uses current model rates; the stored estimate preserves the rates assigned to the recorded usage.
Neither figure reconciles your provider invoice, subscription, credits, or negotiated prices. A change in current rates can change an estimate without changing the underlying usage.
Coverage describes how much captured usage can be priced. Partial estimates include only the charges supported by available token details and rates. Unpriced usage is unknown, not free. Older records may use disclosed cache assumptions. Even full pricing coverage does not prove every device, tool, or provider charge was captured.
Where usage went
| View | What it helps you inspect |
|---|---|
| Project | Estimated cost grouped by assigned work area; unassigned usage remains visible. In team scope, the grouping uses team projects or shared repositories. |
| Model | Which models account for the captured usage and estimated cost. |
| Day | When usage occurred, grouped by UTC day. |
| Pull requests, personal scope | Usage associated with a uniquely matched, recorded merged GitHub pull request: a proposed change reviewed together. |
| Branches, personal scope | Usage allocated to commits grouped by their recorded branch and repository. |
These are alternative views of the same record; do not add their totals together. Pull-request and branch figures use usage dates in the selected period. They are not lifetime costs of those changes. A pull request can include work outside that period, other people's work, and usage Centrail never captured. Missing or ambiguous matches remain separate rather than being charged to every possible pull request. Branch names alone do not establish pull-request membership.
Linked commits and allocated cost per commit
A linked commit is a recorded Git change associated with usage. It does not prove an agent caused that change or that it reached production.
Allocated cost per commit divides the estimated cost assigned to commits by the distinct commits receiving that allocation. Unattributed usage and usage in abandoned groups are kept outside that average. Missing pricing or an unsupported filter combination can make the average unavailable; that is different from zero.
A lower average can reflect smaller commits, a different task, model changes, or different capture coverage. It does not establish better quality, greater productivity, savings, or business value. Use it to investigate comparable work, alongside the usage that remains unmatched.
Team numbers include only eligible shared usage, and some team work metrics cannot follow every filter. Read the explanation beside an unavailable number and how team numbers work.
Environmental footprint
Energy, water, and CO₂ are separate estimates derived from tokens and model assumptions. They are not metered consumption or utility charges. The methodology explains the sources, assumptions, and ranges.