Your agent can use any model in the Rack
A new rack tool lets the agent call the model you assigned to a Rack area: text, vision, transcription, embeddings, rankings, images and System One.
Whatever model you put in a Rack area, the agent can now call it on purpose in the middle of its work. It asks the Rack what is assigned, then sends text or files to the right area.
Text, vision, transcription, embeddings, similarity, reranking, text to image and the System One areas run today. An area with no runner yet says which model and what is missing, and an empty area says how to assign one.
Local models are free; provider models count in your usage like any other spend. Each call shows as a card in the chat with the model, the time and the result. Voice prompts and dictation keep their own path.