Start with the 50-second overview above: what changes when your tools point at one address, and how each request then costs less. It has captions and plays without sound, and the numbers in it are an illustrative example, not a customer result.
This is a recorded walkthrough, not a live event. It follows one developer on their own machine, moving from a shared Claude login to a Proxen key issued in their name, and then making the same change in Cursor and in a plain SDK script. Total change per tool: one base URL and one key.
What the walkthrough covers
- Claude Code. Setting the base URL and auth token environment variables, starting a session, and confirming the first request shows up under the developer's name in the spend view. Claude Code officially supports routing through an organization's gateway, and the walkthrough shows the supported configuration.
- Cursor. Entering the Proxen endpoint and key in the editor settings, saving one profile per allowed model, and switching between them.
- A script. Overriding the base URL in the official SDK, giving the script its own service key, and a nightly job written to collect its results later, so it can use the 50 percent batch rate (OpenAI Batch API and Anthropic Message Batches docs).
- Hitting a cap. What the developer sees when they reach their monthly cap, and what the team lead does to raise it.
Who it is for
Developers who will make the change on their own machines, and the team lead who will answer their questions on the day.
Read the written version in Pointing Claude Code, Codex and Cursor at one gateway, or request the recording.