Token Accounting
Account for every token your agents spend. Attribute cost to the code it produced, see spend per session and per PR, and find where tokens are wasted.
Monthly spend trends
Watch spend month over month — the aggregate line and a breakdown by model — so you can see what's climbing and which models are driving it.
Tokens by pull request
See the cost of every agent session that helped plan or write the code in a pull request, rolled up into the true cost of shipping it.
Add retry & backoff to the webhook dispatcher
- Planned the retry & backoff strategy$4.10Claude Code · Opus 4.8 · 1.2M
- Implemented the dispatcher changes$5.80Claude Code · Sonnet 4.6 · 3.4M
- Wrote the integration tests$1.90Cursor · Sonnet 4.6 · 1.1M
- Addressed review comments$3.20Codex Cloud · GPT-5.5 · 0.8M
Tokens by developer and team
Cross-agent token budgets and trends. Every agent a developer uses rolls into one bar, segmented by model, against the budget you set.
- ACAva Chen$580 / $640+12%
- PNPriya Nairover$700 / $560+21%
- SKSara Kim$560 / $640+4%
- MLMarcus Lee$520 / $560-6%
- DSDiego Santos$330 / $480-14%
Token efficacy
The metrics that tell you whether your spend is working: the average cost of a merged PR and how much code gets generated for every line that ships.
Across 312 merged PRs in the last 30 days — every agent session that planned or wrote the code, summed and divided.
Agents generate ~24 tokens of code for every one that survives to production. The leaner the ratio, the more of every token you keep.
Pick your targets — the teams and repos where cost-per-PR and generated:production run highest are where harness investment pays off most.
Find where tokens are wasted
Roughly 38% of spend goes to sessions on branches that never merge — retries, dead ends, and abandoned work. Surface them and reclaim the budget.