No infra. No card-per-provider. Just one wallet and an OpenAI-compatible endpoint — with every frontier model on tap.
30 seconds. No card required.
Sign in with Google in one click, or use email and password. We create a wallet, a default project, and a starter API key automatically. Everything is scoped to your account from the first second — no shared sandbox, no throwaway keys.
UPI, cards, netbanking.
Add funds through Razorpay. The moment payment clears, your balance is live — no manual approvals, no batch settlement delays. Your one wallet funds every model and every modality.
Scoped, hashed, revocable.
Create as many keys as you need — one per project, per environment, per teammate. Each key can be scoped to specific models and given a spend cap. Keys are hashed with a server-side pepper before storage, so even we can't read them back.
One line change. Zero refactor.
TokenBaazar speaks the OpenAI Chat Completions wire format bit-for-bit. Use the official openai-python or openai-node SDK, LangChain, LlamaIndex, Vercel AI SDK, Cursor, or any tool that already targets OpenAI. Change the base URL, paste your key, done.
The only difference from a plain OpenAI call is the base_url. Everything else — models, messages, streaming, tools — is identical.
from openai import OpenAI
client = OpenAI(
api_key="sk-tb-YOUR_KEY",
base_url="https://tokenbaazar.in/api/public/v1",
)
resp = client.chat.completions.create(
model="openai/gpt-5.4-mini",
messages=[{"role": "user", "content": "Hello!"}],
)
print(resp.choices[0].message.content)Your request hits our edge, gets authenticated and metered in under 40ms, and is forwarded straight to the upstream provider. Streaming tokens pass through as they arrive — we never buffer or replay your payloads.
The moment the upstream response is complete, we calculate cost = (input_tokens × input_rate + output_tokens × output_rate) and debit your wallet atomically. The exact numbers appear on your logs within seconds.
Your key is hashed with a server-side pepper and compared against the stored hash. No plaintext keys ever touch disk. Revoked keys stop working within seconds across every edge node.
Transient upstream 5xx errors are retried with exponential backoff. If a request fails after retries, you are not charged — we only debit the wallet on a successful completion.
We aggregate demand across thousands of developers and buy tokens at wholesale. You get the same models, same providers, same wire format — at 40% of the sticker price. Every request is billed on the exact tokens the upstream returns, not on an estimate.
Chatbots, copilots, and autonomous agents that need reliable access to frontier chat models with tool calling and streaming.
Products that mix chat, image generation, TTS, and transcription — one key, one bill, one dashboard.
Real-time interfaces where first-token latency matters. Our pass-through streaming preserves upstream TTFT.
Overnight enrichment jobs, embeddings backfills, and offline evals — pay per token, no reservation.