AgentsRoom Default reasoning effort: one level per CLI, every new agent starts there
A real session, filmed in the app: a new agent shows Default under Reasoning, so the level used to be set agent by agent. In Settings > AI providers, Claude Code is put on Very high and Kimi Code on Low, each CLI offering only the levels it accepts. Back in the agent window, the Reasoning card still says Default with Very high inherited under it, and the agent boots at xhigh effort without its settings ever being opened.
Read the full feature pageTranscript
Set one thinking level per CLI. Every new agent starts there. This is AgentsRoom, the desktop app where your AI coding agents work side by side.
When you add an agent, its reasoning effort says Default. Until now, that level was something you set agent by agent. Open the settings, AI providers, and pick a CLI.
Each one has its own Default reasoning effort. I put Claude Code on Very high. Every Claude agent with no effort of its own now starts there, on every project.
Another CLI, another list. Kimi only offers the levels it actually accepts: Default, Low, High and Max. I keep that one on Low.
Small jobs, cheap runs, no thinking I did not ask for. Back in the agent window. The reasoning card still says Default, and under it, the level it now inherits: Very high.
I add the agent, and it boots with xhigh effort, written in its own terminal. I never opened its settings. One level per CLI, and every agent follows it.
Default reasoning effort is in AgentsRoom. The link is in the description.