Current, there are at least two parallel mechanisms that contain information of the external backend states:
- Explicit setting a ref() value
- Detecting an actual running instance
- ...
This leads to various problems, e.g. in case a backend crashed (or failed to start) the state might not be up to date.
Also, this leads to unclean code paths, e.g. when backends should intentionally be restarted to free up resources (for editing images via tool call).
Goal would be to get rid of any manual state setting and have everything "detected" automatically. Find a clean architecture that allows that.
Current, there are at least two parallel mechanisms that contain information of the external backend states:
This leads to various problems, e.g. in case a backend crashed (or failed to start) the state might not be up to date.
Also, this leads to unclean code paths, e.g. when backends should intentionally be restarted to free up resources (for editing images via tool call).
Goal would be to get rid of any manual state setting and have everything "detected" automatically. Find a clean architecture that allows that.