Willow "Apps", ie user-defined modules and advanced workflows

[doebi]

I'd like to start a discussion on the latest status of the "apps" idea in the willow documentation, since I am seeking to build something similar for myself.

From heywillow.io:

Chaining of Functions (Apps)
We may support user-defined modules to chain together any number of supported tasks within one request, enabling such things as:

Speech in -> LLM -> Speech out

Speech in -> Arbitrary API -> Speech out

...and more, directly in WIS!

• Could anyone share their take on this?
• What is currently possible?
• What would you recommend?
• How would you approach this?

Looking at the flow chart in the docs, I guess currently it's only possible to invoke API calls from clients, such as the willow on ESP, web rtc or the ts client library. If I want to run multiple steps and API calls, the end client needs to sequentially call these APIs, as there is no possibility of a direct machine to machine comunication. While this might work, this heavily increases response times.

As a first step I'd like to see WIS to have a webhook feature, where it sends the JSON response of any TTS task. In my case I'd send this to a node-red server, where I can build custom workflows.