Model id, like 'google/gemma-4-31b-it'.
Model calls per turn: 20 for chat, 8 or 12 for workers.
Context budget: 32000 for chat, 16000 for workers.
Upper bound on generated tokens per call, 8000.
Sampling temperature, 0.3.
OptionalchatChat-only mode. When it is active the model gets no tools, and the user sees its notice once. Without it the session always offers the tools.
Everything a session needs.