Inference
An OpenAI-compatible API that bills like the rest of your account.
Open-weight models behind one endpoint and one key. Per-token rates you can read before you call, drawn from the same balance as your compute.
Nothing to port.
The endpoint speaks the OpenAI wire format, so migrating is a base URL and a key. If your code already talks to a hosted model, it already talks to this.
- Your existing client works
- Point the OpenAI SDK at our base URL and pass a key. Chat completions, streaming and tool calls behave as the client already expects.
- One key, scoped
- A key can be restricted to named models, so a service that should only reach a small model cannot spend on a large one.
- One balance
- Tokens draw down the same credit balance as compute. There is no separate inference invoice to reconcile.
- Priced per model, in your currency
- Every model in the catalog carries a published rate. You see the cost of a call in the currency you were quoted, not a converted estimate.
Python
from openai import OpenAI
client = OpenAI(
base_url="https://api.alynrex.com/v1",
api_key=ALYNREX_KEY,
)
client.chat.completions.create(
model="<from your catalog>",
messages=[{"role": "user", "content": "Hello"}],
)The model list is set by an operator and changes, so it is not printed here. The console shows what is currently served, what it costs and where it runs.
Every model says where it runs.
A model served from inside Indonesia is marked as such in the catalog, as an explicit assertion recorded against that model rather than something inferred from its name or its provider. Models served from outside are marked too. You should not have to guess, and you should not have to ask.
Read the sovereign AI brief