Introduction

Paranine (P/9) puts leading open-weight models behind the API your code already speaks. Point an OpenAI SDK at Paranine's address, use a Paranine key and pick a model. Prices are in rupees per million tokens, paid from a prepaid balance: no foreign card and no exchange-rate surprise at the end of the month.

https://sovereign.paranine.com/v1

What you get

• OpenAI-compatible. Chat completions, streaming, tool calling and embeddings in the OpenAI request and response shape. Most code changes two lines.
• Rupee billing. Prices per million tokens in PKR, a prepaid balance, and the cost of every request in your logs.
• Keys with limits. Give each key an expiry, a credit cap, a list of allowed models and its own rate limit.
• Visibility. Latency, time to first token, tokens and cost for every request, by key, project and model.
• Residency. Choose models that run inside Pakistan when your data must stay in the country.

Where to start

1. Quickstart: a key and a first request in five minutes.
2. Models: every model, its id and its price.
3. Chat completions: the request and response in full.
4. Errors and retries: what can go wrong, and what to do about it.

Moving from OpenAI? Migrating from OpenAI lists the few differences.