Ollama · LM Studio · vLLM

Driven as

A headless process, one per card, in its own isolated workspace. Deskworth never holds a key for it: the subscription and the credentials stay yours.

What the desk reads back

The raw stream of the process, turned into a trace, one line per phase, per tool, per process. Never a summary, and the raw output stays open underneath.

Where it comes from

This entry is not written by hand. It is read from the core’s provider registry, so a fourth provider appears here the day the core supports it, and no provider can appear here without existing there.

Registry: local-openaiHow an adapter drives it.