CodeSpend pulls usage and seat costs from every AI coding tool your team pays for, matches them to developers and merged pull requests, and shows where the money is wasted.
Read-only admin keys. We never read prompts or code.
Usage-based pricing means one long agent session can cost more than a month of seats. Most teams find out on the invoice.
Uber reportedly used up its full-year 2026 AI coding budget by April, then capped spend per engineer per tool.
DesignRush, 2026After GitHub Copilot moved to metered AI Credits in June 2026, one Pro+ user reported losing 8% of a monthly allowance in two hours.
gHacks, June 2026Gartner predicts AI coding costs will surpass the average developer’s salary by 2028 as token use climbs.
Gartner, June 2026Move the sliders to match your team. The math runs in your browser; nothing is sent until you sign up.
An estimate from the assumptions above, not a promise. CodeSpend replaces them with your real numbers.
Get the real number for my teamSetup takes about 15 minutes for an admin.
Read-only keys for Anthropic, Cursor, GitHub and OpenAI. Claude Pro/Max users connect through Claude Code’s built-in telemetry.
Costs are tied to each developer and their merged PRs, so you see cost per PR by person and team.
Idle seats, plans that would be cheaper flat-rate, premium-model overuse and teams heading over budget.
Each covers part of the picture.
| Option | What it sees | What it misses |
|---|---|---|
| Personal trackers (ccusage, CodexBar) | One developer’s own limits and usage | No team view, no budgets, no link to output |
| API gateways (Portkey, LiteLLM) | Traffic routed through your API keys | Seat plans, Cursor, Copilot, Claude subscriptions |
| Each vendor’s admin console | That vendor’s spend | The other tools, and what the spend produced |
| CodeSpend | Claude Code, Cursor, Copilot and Codex together, including subscription users | Early product: four integrations so far |
Free for early teams while we build it with you. Tell us a little about your setup.