India's AI Gateway
Token Sutra
Route every model through one key. Watch tokens, cost, and latency in one place — for Hindi, Tamil, English, and whatever you ship next.
One API
Drop-in OpenAI-compatible endpoint
One key
Issue, rotate, revoke in the dashboard
One bill
Tokens and spend in INR, per key and per model
India-first
Indic scripts, regional latency, local teams
This is the product. Login and it is already yours.
Explain a token in one line.
A token is the smallest chunk of text a model reads or writes — usually a few characters.
This is the product. Login and it is already yours.
Routing
A single thread through every model
One base URL. Swap models without rewriting clients. Fail over between providers, split traffic, and pin versions — from config, not code.
Keys
Keys that behave like production
Named keys, environment split, instant revoke. Secrets are shown once and stored hashed — never printed twice, never sitting in a log.
Shown once. Stored hashed.
Observability
Tokens you can actually see
Prompt vs completion, cost, latency, per-model mix. Every request is accounted for, in INR, before your invoice ever surprises you.
India
India in the request path
Indic tokenization, regional routing, and billing that assumes India — not a US console with a flag emoji bolted on.
नमस्ते
If you already speak OpenAI, you already speak Token Sutra.
Point the official OpenAI SDK at the gateway. Same calls, same shapes — one key across every model.
import OpenAI from "openai"; const sutra = new OpenAI({ apiKey: process.env.TOKEN_SUTRA_KEY, baseURL: "https://api.tokensutra.in/v1",}); const res = await sutra.chat.completions.create({ model: "sarvam/sarvam-30b", messages: [{ role: "user", content: "नमस्ते, एक लाइन में समझाओ।" }],});India
A gateway that does not treat India as a locale toggle.
Ten scripts, one prompt, one thread. Indian-language tokenizers elsewhere waste tokens; the gateway makes that waste visible and optional to avoid.
- Hindiनमस्ते, एक लाइन में समझाओ।
- Tamilவணக்கம், ஒரு வரியில் விளக்குங்கள்.
- Teluguనమస్తే, ఒక్క లైన్లో చెప్పండి.
- Bengaliনমস্কার, এক লাইনে বলুন।
- Marathiनमस्कार, एका ओळीत सांगा.
- Kannadaನಮಸ್ಕಾರ, ಒಂದೇ ಸಾಲಿನಲ್ಲಿ ಹೇಳಿ.
- Malayalamനമസ്കാരം, ഒരു വരിയിൽ പറയൂ.
- Gujaratiનમસ્તે, એક લીટીમાં કહો.
- Punjabiਸਤ ਸ੍ਰੀ ਅਕਾਲ, ਇੱਕ ਲਾਈਨ ਵਿੱਚ ਦੱਸੋ।
- EnglishHello, explain in one line.
From zero to observed in four steps
01
Create an account
Sign in to Token Sutra.
02
Mint a key
Name it, scope it, copy it once.
03
Point your client at the gateway
Change one base URL.
04
Watch usage land
Tokens, cost, and latency in the dashboard.
Quiet where it counts
- Keys stored hashed; secrets shown once
- Per-key scoping — rotate and revoke instantly
- HTTPS only, everywhere
- Usage and spend visible per key and per model