Coding-agent costs explained

How Claude Code, Cursor, Copilot, Codex, Devin and the other coding tools bill, why their dashboards disagree with your invoice, and how to get one number per engineer and team.

By Keyveve, Inc. · Updated 2026-10-25

A team that uses several coding agents gets several bills, each in its own unit, at list price, with its own dashboard. Getting from those to one number per engineer and team takes four steps: read each tool's usage per person, decide who each person is, price it at what you actually pay, and add it up by team.

The three ways coding tools bill

• Seats. A fixed price per person per month, whether they use the tool or not. A seat nobody touched for a month is money you can save at the next renewal.
• Credits or compute units. Many agents meter work in their own unit: credits, ACUs, or requests to premium models. The unit's price comes from your contract, so the tool's dashboard can't show it in dollars on its own.
• Tokens. Agents that run on a model provider's API, such as Claude Code with an API key or OpenCode and Aider on your own keys, bill per token at the provider's rates, which change with caching and the model used.

Most tools mix these: a seat that includes some usage, then credits or tokens past it. That is why a tool's dashboard, which shows list price, rarely matches the invoice.

One person, many names

Each tool knows people by its own name for them: an email, a login, a user id, or the name of the API key they used. Before costs can be added up, each of those names has to point to one person. Email does most of the work; the rest needs matching once, and a key's name or owner ties agents on API keys to the person who runs them.

Price it at your contract, day by day

Usage should be priced at the contract in force on the day it happened: the seat price, the price per credit, a discount on the tool's own dollars, and the usage the seats include. A renewal in the middle of a month then splits correctly, and the total can be checked against the invoice line by line.

Add it up by team

People move between teams, so a team's cost is each person's cost for the days they spent in it. Teams usually come from the directory, with an HR export for cost centers. With that, a coding-agent bill becomes something finance can charge back, and a manager can see their own team's spend.

Keep it in check

• Limits inside the tools. Some tools take a monthly limit per person by API, so the tool itself stops someone at it. For the rest, an alert at 80% of a person's total across all their tools works.
• Unusual days. A day five times someone's usual one is how a stolen seat, a leaked token or an agent left looping shows up, and it's worth hearing about the day it happens.
• Idle seats. Seats unused for 30 days, and seats still held by people who left, are the quickest saving.

With Veqen

Veqen does each of these steps across Claude Code, Cursor, GitHub Copilot, Codex, Devin, Windsurf and the rest: by API with read-only admin keys, from exports for tools without an API, and from the provider keys agents run on. See work and coding agents and the docs. To try it offline first, npx @veqen/sdk coding-bill cursor.csv copilot.csv --teams teams.csv turns your exports into a bill per team without uploading anything.