OpenAI-compatible · text, image, speech · billed in INR
One Key. Every Kind Of Intelligence.
Chat, image generation, speech and transcription behind a single OpenAI-compatible endpoint. Point your SDK at ai.upfyn.com/v1, keep the code you already have, and get one credit balance and one invoice — in rupees.
5,000 free credits · no card · a working key in under a minute
What One Key Covers
Four shipped routes on one balance. Everything is billed from the same credits, so there is no second account to reconcile and no per-modality invoice.
Chat
/v1/chat/completions
Streaming, tool calls and structured outputs, in the shape the OpenAI SDK already sends. Server-held multi-turn sessions if you would rather not keep a conversation store.
Images
/v1/images/generations
Synchronous, or queue it and poll — the async variant returns a task id you can collect later.
Speech
/v1/audio/speech
Text to audio on the same key and the same credit balance, sync or async.
Transcription
/v1/audio/transcriptions
Audio back to text, billed from the same balance as everything else.
One Diff. Same Code. New Gateway.
Two domains swapped. Everything else stays.
base_url="https://api.openai.com/v1"
base_url="https://ai.upfyn.com/v1"
- Same OpenAI SDK. No rewrite. No new client.
- Bala, Yuva and Rishi stay put. Providers and prices change underneath them; the id in your code does not.
- Every model has multiple backends behind it. A provider outage is the same id on a different route.
One Balance, Every Modality.
Reservations and refunds
A failed request costs nothing. Credits are reserved, then released if the call does not complete.
Idempotent billing
Payment events survive duplicate replay. The duplicate-event paths are covered by tests, so a webhook delivered four times still lands on the balance once.
Diagnostics
A preflight check at /v1/diagnostics/run tests every model your account can reach. Wire it into CI to catch broken provider config before launch day.
Usage you can query
The ledger and usage endpoints are part of the API, not a dashboard-only feature. Attribute spend yourself at /v1/me/ledger and /v1/me/usage.
Let Your Users Pay For Their Own Inference.
Their usage bills to them — not to you. Scoped tokens, real consent, no key sharing.
- OAuth 2.0
- Scoped tokens
- Consent screen
- Per-user billing
- MCP manifest
- No key sharing
One key. Every kind of intelligence.
5,000 free credits, no card, and a working API key in under a minute. The same models run the desktop app, so anything you build on the API behaves the way the product does.