Could we get some kind of "dummy" Realtime API for testing purposes?

I’ve had an incredibly fun idea for a game that would use the Realtime API and spent a while messing about in the Playground and it works very well—

But I’m a shit programmer. Implementing it in a game engine is going to take me a while, and one thing holding me back is… the price. There’s going to be a lot of testing, and I can tell I’m probably going to lose at least $50 just from the testing sessions where I’ll repeatedly check if different aspects of the implementation works.

I wish there were a “dummy” API that didn’t do any generating, that we could use to test our implementation without wasting money/tokens/compute. Something that just sends dummy responses based on specific requests so that once everything is working, we can change one parameter (like the model?) and immediately switch to the regular API.

Come to think of it, the Chat Completion API already has that in a way, since we can use 4o-mini for cheap testing and switch to 4o or o1 once everything is set up. In its current state, the Realtime API doesn’t have anything similar, which is making it needlessly expensive to implement.

I don’t know, is this a terrible idea?

1 Like

Forgive me if I’m oversimplifying the problem. However, if one needs canned / dummy responses, why not setup a local instance of FastAPI that just takes all the same parameters as the 4o model and replies with preset responses? This way you could still accomplish input validation, error handling, etc. (Another variant of this idea would be to integrate with a free model online – e.g. HuggingFace – and then switch the model after testing. This does assume you could live with longer latencies in testing with a free model, however.)

As you’ve mentioned, one could always use the mini models to minimize testing costs.

2 Likes