Router
The Router is a single OpenAI-compatible endpoint in front of every model the platform can reach. Your application calls one URL with one key and names the model it wants, so changing model becomes a string change rather than a new integration.
Because every call passes through the platform, cost and latency are recorded against the model definitions in this catalog with no instrumentation on your side. Agents use the Router as well: the primary and fallback models on an agent's Connections step are chosen from the same catalog.
The model catalog
LLMs lists every model reachable through the Router, with the Model ID your calls use, the upstream provider, the region, what the model is suited to, its capabilities and context window, and its pricing.
These definitions are maintained by InteractiveAI, so pricing and tokenization stay current as providers change their rates and you never configure either.

The same model can appear once per region, so filter by region when it matters where a request is served.

Opening a model shows its match pattern, the expression the platform uses to recognise it in your traces, along with its tokenizer and full pricing per million units. Below that, Model observations lists every generation that used the model, with the same filters and export controls as Observability. That is where you see how a model actually performs in your project rather than how it is advertised.

Router keys
A Router key is what your application authenticates with. It is shown once when created, so copy it then, and it is sent as a bearer token to the Router's chat completions endpoint. The dialog that creates it includes a ready-to-run request so you can confirm the key works before wiring it into anything.
Each key carries a spend limit and a reset period, and the list shows how much of the limit remains. That makes a key a unit of budget control as well as authentication: give an application or a team its own key and cap what it can spend. Agents get their own key generated when you deploy them.

Last updated
Was this helpful?

