OpenAI and Anthropic formats
POST /v1/chat/completions follows the OpenAI format and POST /v1/messages follows the Anthropic format, both with streaming. Just change the base URL and API key.
Product
MarkasAI: pay-per-token LLM API in Rupiah for Indonesian developers
MarkasAI is a MarkasDev product: one API key for many models, compatible with the OpenAI and Anthropic formats, billed per token from a prepaid Rupiah balance you top up with QRIS, virtual account, or e-wallet.
The problem it solves
Developers in Indonesia who want to use large language models (LLMs) in their apps or coding tools often have to manage separate accounts and bills for each model provider, paid in a foreign currency. Costs also creep up quietly when a tool or script keeps calling the API.
MarkasAI brings models from many providers, such as OpenAI, Google, Anthropic, Qwen, DeepSeek, and xAI, together behind one API key. You pay in Rupiah from a prepaid balance with no subscription, and only for the tokens you actually use. Every key can have its own limits, so the bill never surprises you.
How it works
From balance to first call takes three steps.
- 1
Top up
Top up with QRIS, virtual account, or e-wallet, with payments processed by DOKU. Your balance updates automatically once payment is confirmed.
- 2
Create an API key
Set a spend limit in Rupiah, a per-minute request limit, an expiry date, and allowed models. The key is shown only once, so store it in an environment variable.
- 3
Call the API
Point your SDK or tool at the base URL https://api.markas.ai.id/v1 (Anthropic-format tools use https://api.markas.ai.id without /v1), pick a model ID from the Models page, and go. There is no custom SDK to install.
Features
Responses API, images, and video
POST /v1/responses for Codex CLI and the OpenAI Responses SDK. Images via /v1/images/generations, billed per image or per token depending on the model. Video via /v1/videos/generations, billed per second at the chosen resolution.
Works with the tools you already use
Works with tools that support the OpenAI or Anthropic API, such as Cursor, Claude Code, Cline, Roo Code, Kilo Code, Continue, Aider, OpenCode, OpenClaw, Zed, Open WebUI, and Codex CLI.
Per-key limits
Every key has its own spend limit, requests-per-minute limit, expiry date, and allowed models, so a wasteful tool cannot drain your balance.
Usage reporting
See cost per day, per model, and per key, and download a CSV for your books.
Ready-to-use chat app
Try models without writing code, using the same account and balance. Conversation history is kept in your account and can be deleted at any time.
Transparent billing
A prepaid balance in Rupiah, with no subscription and payments through DOKU. Before a request runs, the highest possible cost of that request is held from your balance as a guarantee. The model then answers, and streaming is passed through as is. When it finishes, only the real cost is taken from the hold, and the rest returns instantly.
Errors before the first byte (502, 503, 504) are never billed. If your balance is too low, the request is rejected with a 402 before it is forwarded to the model. Prompt and response content is not stored in billing logs. Sale prices in Rupiah per 1 million tokens are shown on the Models page at markas.ai.id and in GET /v1/models, and a change applies to later requests.
Who MarkasAI is for
- Developers who use AI coding tools such as Cursor, Claude Code, Cline, Codex CLI, or Aider and want to pay in Rupiah.
- Teams building LLM, image, or video features on their servers with the OpenAI SDK or the Anthropic format who want one key for many models.
- Anyone who needs to cap cost per key and track usage per day, per model, and per key.
- Not a fit yet if you need embeddings, audio, fine-tuning, batch, or file uploads, since these are not available. Calling directly from a browser is also unsupported, so call it from your server.
Quick FAQ
Is it compatible with the OpenAI and Anthropic SDKs?
Yes. /v1/chat/completions follows the OpenAI format and /v1/messages follows the Anthropic format, both with streaming. Just change the base URL and API key.
What does each model cost?
Prices in Rupiah per 1 million tokens are shown on the Models page at markas.ai.id and in GET /v1/models. Image models are billed per image or per token, video models per second. We do not list prices here because they can change.
What happens if my balance is too low?
The request is rejected with a 402 before it is forwarded to the model, so you are not billed for an answer that was never sent.
Are my prompts stored?
Not in MarkasAI's logs. Only metadata (model, token counts, cost, status, time) is recorded for billing. Requests are forwarded to the model provider, which has its own policy. The exception is the chat app, which stores conversation history so you can open it from any device and delete it at any time.
How do I top up?
From the Balance page in the dashboard. Available methods are listed there, and your balance updates automatically once payment is confirmed.
Start with one API key
Sign up at markas.ai.id, top up, and create your first key.