A token-management system for LLM and OCR calls, already running in production inside our own team — it estimates cost before firing, caps budget per key, guards against surprise bills, and was designed with PDPA privacy principles from day one.
It's not just forwarding a request to the model — every call passes through this cost- and privacy-control layer first.
Thai text costs roughly 2x more tokens than English (≈3.8 characters/token for English vs. ≈1.8 characters/token for Thai). We calculate the Thai-language proportion of your text and estimate cost before the real model call — no LLM call wasted on guessing.
Meters usage per API key — pages, tokens, cost, and which model was called. Set a monthly quota and a daily budget cap; going over the cap stops the call before it's spent, instead of finding out when the bill arrives.
Token-bucket rate limiting per tier, with a clear 429 response and Retry-After header — prevents excessive call rates and runaway cost.
Starts with the lowest-cost OCR model, escalating to a pricier model only when needed — and converts a full document page to markdown.
Public holidays, corporate lookups from the Department of Business Development (DBD), and PromptPay QR generation — served from a data lane that never calls a model, at $0 cost, always with the source cited.
The usage-metering system is built following privacy-by-design principles, aligned with the PDPA.
Our flagship product already runs on 1intentAI as its AI/OCR usage-metering layer in real day-to-day operation — before we open it to outside customers.
Issuing an API key today is high-touch and direct — our team will set up quota and budget to fit how you plan to use it.
Contact for API keySee our privacy policy at /en/privacy