OpenLuxWhat is an LLM API aggregator?
An aggregator puts GPT, Claude, Gemini, DeepSeek and more behind one OpenAI-compatible endpoint — a single key and a single bill. Here is how they work, what to check, and how to get started.
Four steps to your first aggregated call
Create an API key
Register in the console and create a key (sk-openlux-). Keys can be rotated or revoked at any time.
Point your SDK at the gateway
Set base_url to https://openlux.lol/v1 — every OpenAI-compatible SDK or client works unchanged.
Choose a model per request
Set the model field to any catalogue name: gpt-6-astra, claude-opus-5, gemini-3.8-flash and so on. Switch freely from call to call.
Check the bill
Every call, its token usage and its cost appear in the console, filterable by model and by day.
What an LLM API aggregator is
An LLM API aggregator is a gateway that sits between your application and multiple model providers. You get one OpenAI-compatible endpoint, one API key and one consolidated bill; the aggregator handles per-vendor authentication, routing and metering behind the scenes.
The pattern is the same one that made OpenRouter popular: instead of integrating each vendor separately, you integrate once and choose models per request by changing a single field.
Why teams use one
Fewer integrations: one SDK and one authentication scheme replace a growing list of vendor clients. One invoice: cost reporting and top-ups happen in one place, in one currency. Instant model choice: new models are typically available the day they ship, without new contracts or accounts.
Operational leverage matters too: routing, failover and usage reporting are handled by the gateway, and access does not depend on where each individual vendor’s API happens to be reachable.
What to check before you commit
Coverage — the models you actually run, including image, audio and retrieval models if your product uses them. Transparency — published unit prices per model, cache pricing labelled separately, no hidden multipliers. Protocol completeness — streaming, tool calling, structured output and multimodal input must behave like the official APIs.
Then the operational questions: how usage is metered and reported, which currencies and payment methods are supported, and how support works when something breaks at 2 a.m.
How OpenLux works as your aggregator
OpenLux exposes 19 production models through one OpenAI-compatible endpoint at https://openlux.lol/v1: text, image, audio and video, plus embedding and rerank models for retrieval workloads. Every model’s price is published in the catalogue, and that price is what your account is charged — metered per request or per token in USD.
The console shows the cost of every call, filterable by model and by day; metering works the same whichever model you pick. Support runs through support@openlux.lol.
Popular models by workload (live pricing)
| Model | Vendor | Price | Vs. official |
|---|---|---|---|
| gpt-6-astra | OpenAI | $0.3677 / 1M tokens | -96% |
| claude-opus-5 | Anthropic | $0.4412 / 1M tokens | -91% |
| gemini-3.8-flash | $0.0551 / 1M tokens | -92% | |
| deepseek-v4 | DeepSeek | $0.0613 / 1M tokens | -90% |
Prices are metered in USD; the full list lives in the model catalogue.
Frequently asked questions
Is an aggregator the same as a proxy?
Conceptually close, but an aggregator adds a model catalogue, unified metering and billing, and usage reporting on top of routing. You can self-host a gateway like LiteLLM, or use a hosted aggregator — hosted removes the ops work.
Do I need separate vendor accounts?
No. One OpenLux key covers every model in the catalogue; there is no onboarding with individual vendors.
Does an extra hop add latency?
There is one added hop, but modern gateways keep it minimal — regional reachability usually matters more than hop count in practice.
Can I move off an aggregator later?
Yes. Because the API is OpenAI-compatible, going back to a direct vendor integration — or to another gateway — is the same two-line change in reverse.
One key for every model
Create a key and make your first aggregated call in minutes.