FAQ
Including the ones with answers we would rather were different. Anything not built yet is listed as not built, not as coming soon.
Yes. Every top-up produces a full Thai tax invoice (ใบกำกับภาษี) in THB from I C Develop Co., Ltd., a company registered in Thailand. That is the reason this gateway exists: paying OpenAI, Anthropic or Google directly gets you a foreign receipt that your accounting team cannot file the same way.
Not for what you spend here. Buying from IC Lueur Ai is a domestic purchase from a Thai company, so it is claimable at 100% as an ordinary expense with no foreign-service VAT self-assessment (ภ.พ.36) and no withholding remittance for a foreign payee (ภ.ง.ด.54). Paying a foreign AI provider directly is what creates those filings. This is a description of how the purchase is structured, not tax advice — your accountant signs the return, not us.
PromptPay from a Thai bank account, minimum ฿300 per top-up. No foreign credit card is needed and there is no FX spread applied at top-up. Credit is prepaid and does not expire on a timer.
No. There is one prepaid balance and you are billed from measured usage — input, output and cached tokens are priced separately because the upstream providers price them separately. There are no tiers and no per-seat charge.
Nothing you already spent changes. The rate in force when a request ran is stored with that request, so a later price change never re-prices past usage. The billing ledger is append-only — corrections are posted as new compensating entries, never by editing history.
In Thailand. Your account, your credit balance, the append-only billing ledger, your usage records and your tax invoices are all held in the AWS Asia Pacific (Thailand) region, ap-southeast-7, along with the gateway and control plane themselves.
No, and nobody else’s does either. No provider — AWS Bedrock, Google Vertex AI, Azure AI Foundry or the others in our catalogue — offers these models from inside Thailand today. Requests are served from the providers’ Asian regions, mainly Singapore (ap-southeast-1) and Tokyo (ap-northeast-1), and some models are only offered on a provider-managed global footprint. Each model’s entry in the catalogue names the regions it is served from, so you can see it per model rather than take our word for it.
Yes. Each API key carries a routing policy that is enforced on every call, not treated as a preference: strict never leaves the nearest region and returns 429 with Retry-After rather than reaching further; regional, the default, allows overflow within the Singapore–Tokyo boundary; available is opt-in and allows anywhere a model is served. See the docs for exactly what each one does.
No. The gateway records token counts, cost, model, provider, region and timestamps — the fields needed to bill you. Prompt and completion bodies are never written to a store or included in any payload that leaves the process, and we do not train anything on your traffic. The upstream provider you route to has its own terms, which we do not override.
No, and that is deliberate. A key is shown once when you create it and only a hash of it is stored, so we genuinely cannot display it again. Create a replacement and revoke the old one. The member portal also has two-factor authentication with recovery codes.
The base URL and the API key. The gateway speaks the OpenAI Chat Completions API, so an existing openai-python, openai-node or LangChain client — anything that lets you set a base URL — keeps working. The one difference worth knowing is the error shape: errors come back as {"detail": "..."} rather than OpenAI’s nested error object.
Yes. Set stream: true and read the response as server-sent events, in the same chunked format OpenAI uses. Note that tokens generated before you disconnect are still billed — an interrupted request is metered like a completed one.
The live catalogue on the pricing page is the answer, and GET /api/models returns the same thing as JSON with no authentication. It spans Anthropic Claude, OpenAI GPT, Google Gemini, Meta Llama, Mistral, DeepSeek and others, across AWS Bedrock, Google Vertex AI, Azure AI Foundry and Fireworks. We do not publish a fixed count here because the catalogue changes.
Yes. An organization can have several people in it with a role each — owner, admin or member — so nobody has to share a password to see the balance or manage keys.
No. We charge provider list cost plus a fixed margin, so per token we are more expensive than a direct contract. If you can hold your own agreement with AWS, Google or Microsoft and handle the foreign-service tax paperwork yourself, that route is always cheaper. What the margin buys is the Thai tax invoice, PromptPay settlement, routing you control and support in your timezone. If price per token is the only thing that matters to you, we are not the right gateway.
Not today. Per-key request limits are planned for a later phase and are not enforced. The only throttling in the system right now protects the member portal’s login form against repeated guesses, and has nothing to do with calls to /v1/chat/completions.
We do not publish any yet, because we do not measure them yet. The status page will carry time-to-first-token and tokens-per-second per model, measured continuously from Bangkok by the same benchmark that sets our routing defaults. Publishing numbers we did not measure would be worse than publishing none, so the page stays empty until the measurements are real.
The docs cover the technical side in more detail, and pricing shows the live per-model rates.
