Home → Guides → What an LLM gateway is, and when you need one

What an LLM gateway is, and when you need one

The word covers three fairly different products. Knowing which one you are looking at is most of the decision.

Three things called a gateway

1. A routing relay. One base URL, many providers behind it, one balance. You change a model string instead of an SDK. This is what APICLAN is.
2. An observability proxy. Sits between your code and your own provider accounts, adds logging, retries, caching and spend limits. You keep your provider contracts and pay for the layer.
3. A self-hosted router. A library or service you run yourself that normalises several provider SDKs into one interface. No third party in the request path, but you operate it.

They solve different problems and the comparison between them is usually pointless until you have decided which problem you have.

What a relay buys you

One integration instead of four. Claude, GPT, Gemini and Grok have different SDKs, different auth, different streaming formats and different error shapes. Behind a compatible relay they are one client and one key, and switching model is a one-word change in a config file.

One balance instead of four accounts. Four provider accounts means four cards on file, four sets of rate limits to track and four invoices to reconcile. For a small team that overhead is real.

Prices that are not the provider's list. Relays that buy capacity differently from a metered API account can publish rates below list. On this platform every model shows its official price beside what you pay, on its own page, and the ratio is fixed rather than negotiated.

What it costs you

• A hop in the request path. One more service that can be down. If you need an uptime commitment in writing, a relay that will not publish one cannot give you it.
• Later access to new models. Providers launch on their own API first.
• No provider-side contract. Data processing agreements, region pinning and enterprise data controls live on the provider account, not on the relay.
• Account-level features do not pass through. Batch pricing, fine-tuning and file storage stay behind.

How to decide

Ask what happens if the answer is wrong. If a compliance officer will eventually ask you to produce a signed agreement covering customer data, go direct and stop reading comparisons — the price difference will not cover that conversation. That case is laid out in full in APICLAN vs paying the provider directly.

If the money is coming out of your own pocket, if the workload is development, internal tooling, evaluation or an agent loop against your own code, then the per-token rate is the whole decision and the arithmetic is public. Most teams end up doing both: direct for the customer-facing path that needs the paperwork, a relay for everything else, which is usually where most of the token count lives and none of the compliance risk does.

Start using it

No subscription, no monthly minimum, no sales call. Top up with USDT and spend what you use — 1 USDT gives you 2 credits of API balance.

Read the 30-second quickstart

Prices quoted on this page are regenerated automatically from live billing data. Third-party terms are quoted from that party's own published documentation.