If you are a backend, ML, or full-stack engineer in Bengaluru, Hyderabad, Pune, or anywhere else in India, you have probably hit the same wall I did back in 2023: OpenAI billing only accepts Visa, Mastercard, American Express, or Discover from a supported country list, and India is not on it. UPI is the default payment rail in India, but OpenAI does not expose it. This article is the production-grade playbook I now use to ship LLM features from India — including drop-in client code, concurrency tuning, pricing math, and the benchmarks I measured on a real workload.

Why Indian developers get blocked at checkout

OpenAI's payment processor rejects most Indian-issued cards because of high chargeback ratios and 3-D Secure gaps. Engineers typically fall back to one of three workarounds:

The relay approach keeps your existing OpenAI SDK code unchanged. You only swap base_url and api_key. I have been running HolySheep AI (an OpenAI-compatible gateway at https://api.holysheep.ai/v1) in production for 11 months across two SaaS products with zero SDK rewrites.

Architecture: what an OpenAI-compatible relay actually does

Conceptually, the gateway is a thin reverse proxy in front of upstream providers (OpenAI, Anthropic, Google, DeepSeek). Your client speaks the OpenAI REST schema; the gateway translates the auth header, applies its own rate limit, and forwards the request. The relevant primitives for an Indian engineering team are: