Getting started
Models
Every model in the catalog is reachable from every protocol. Pass the model id in the request body; the base URL you used does not restrict which models you can ask for.
Finding the id
The catalog page lists every model with its context window and published rate. The same list is available to your key over the API, which is the better source if you are building against it programmatically.
curl https://api.overagent.cc/v1/models \
-H "Authorization: Bearer $OVERAGENT_API_KEY"Fallbacks
Some models run a safety classifier that occasionally refuses a turn that a sibling model will answer without complaint. Rather than hand that refusal back to you, the gateway can retry the same turn on another model.
This is on by default for the models where it matters. You control it with the fallbacks field:
"default"— use the built-in chain for that model. This is the default.- A list of model ids — try these, in order, after a refusal.
[]— never retry. You get the refusal verbatim.
When a fallback answers, the response is explicit about it: the model field names whoever actually replied, and usage.iterations records how many models were tried. You are billed for each attempt, which is why the wallet hold is sized for the whole chain rather than the first hop.
Context windows
Context limits are the model's own. The gateway adds a ceiling of its own on raw prompt size — see Limits — but that ceiling sits well above every model's window, so in practice the model's context is what binds.