Operon v1.4.9

Changed

• ACP agents use their advertised model, mode and reasoning effort options, with support for switching them during a chat.
• Kimi discovers available settings when starting a session; Grok reads each model's supported reasoning effort levels.

Fixed

• The reasoning effort picker updates when the model changes and clears options the agent no longer offers.
• ACP sessions apply the selected reasoning effort before the first prompt and report unsupported values instead of silently ignoring them.