The problem
Coding agents are the heaviest AI spenders in most companies now. They resend the same context on every step, run in loops, and each developer’s usage vanishes into one provider invoice. You cannot see who spent what, cap anyone, or cut the obvious waste.
How it works
Any coding tool that speaks the OpenAI API can point at aigw. Set the base URL and a virtual key, and the traffic flows through the gateway.
export OPENAI_BASE_URL=https://gateway.aigw.app/v1
export OPENAI_API_KEY=aigw_your_key
That is the whole change. Your agent keeps calling the same models. aigw sits in the path and does the rest.
What you get
- A budget on every key, per developer or per team, enforced at the edge.
- Cost attributed to the person, team, and project behind each key, gross and net of what aigw saved.
- Lossless savings: duplicate calls are coalesced, repeated context is read from the cheaper cached tier, and identical requests come straight from cache. None of it changes the output.
- The same policy, secret scanning, and audit log as the rest of your AI traffic.
Full setup, tool by tool, is in the coding quickstart.