Example setup: a 20-person agency
Illustrative example, not a customer. How a 20-person agency with six client accounts could move from four shared logins to one key per person, with AI spend billed back to each client.
Overview
Illustrative example, not a customer. Numbers are the setup values, not results.
A 20-person agency doing design, content and light development for six clients
Four shared logins: two Claude Max seats, one ChatGPT Pro plan, one Gemini plan, passwords in a pinned message
Company API and Team accounts, 20 person keys and 2 service keys, one team per client, budgets per person and per client
Challenge
In this example, the agency's problem is not the bill. It is that nobody can say which client's work used which account, so AI costs cannot be billed back and client code from different accounts sits in one chat history.
The owner also has two contractors who come and go, and each time one leaves the passwords have to be rotated and re-sent to everyone.
setup values
Key per person. Twenty people, two services, zero shared passwords
Teams, one per client, so spend lands against the client it belongs to
Budget alert per person, with a hard stop at 100 percent
Challenge
Illustrative example, not a customer. Four shared logins across six clients means no way to bill AI work back, no confidentiality between client histories, and a password rotation every time a contractor leaves.
The agency opens its own API accounts with Anthropic, OpenAI and Google and a Team plan for the chat apps. Those keys go into Proxen and stay server-side. The console gets six teams, named after the six clients, plus an internal team for the agency's own work.
Every person gets one key. People who work across clients get a key per client team so their spend is attributed correctly; switching is a different profile in Cursor or a different environment variable in a script. Contractors get keys with a shorter model list and a lower cap, and a key is revoked the day a contract ends.
Setup
One key per person, grouped into one team per client. Each client team has its own monthly budget and its own retention setting, so the agency can show each client exactly what was spent on their work.
Budgets are set per person inside each client team's budget. The alert at 80 percent goes to the person and the account lead. At 100 percent the key stops, which in this example is the point at which the account lead decides whether the client's budget should grow rather than discovering it on the invoice.
At month end the spend view, grouped by client team, is the line item on each client's bill. Prompt bodies are not stored for any client team unless that client asks for it in writing. The two Claude Max seats, the ChatGPT Pro plan and the Gemini plan are retired after the first full week.
built on evidence
Every setup example rests on the same four facts about how Proxen works.
Key per person, with its own budget and log
Shared passwords left once rollout is done
Base URL to change in Claude Code, Codex, Cursor or your SDK
Batch discount OpenAI and Anthropic give on work that can wait
Want a setup like this for your team?
Send one month of usage and we will map a setup like this to your accounts, with the saving to expect.
