I built a hard spend cap you can put in front of ChatGPT-compatible agents. Can you try to break it?

I’m the builder of Valta. It’s an OpenAI-compatible proxy. You swap two lines (base URL and key) and each agent run gets a hard cap. Calls over the cap are denied before they reach OpenAI.

llm = ChatOpenAI(
base_url=“https://valta.co/v1”,
api_key=“vk_live_…”, # Valta virtual key, not your OpenAI key
max_tokens=500,
)

What I’d like tested: a LangGraph loop with retries that should trip a per-run cap. Demo repo: GitHub - Valta-hq/langgraph-spend-cap: LangGraph retry loop with a $3 per-run cap. Valta Cap denies the 4th paid call before the provider. recursion_limit does not do this. · GitHub. Setup guide: Set up the OpenAI proxy (8 steps to your first deny) | Valta Docs

Limits today: OpenAI only, /v1/chat/completions only, no streaming. Free plan is 1 agent. If it fails to stop a loop, I want to know.