Overview
Loop supports thinking modes that let you see and control the model’s chain-of-thought reasoning. When enabled, the model shows its reasoning process before providing a response.Thinking Levels
Shift+Tab cycles off → low → medium → high. Settings also accept minimal, xhigh, and max when a model supports them; the keyboard cycle does not include those three.
Controlling Thinking
Keyboard Shortcuts
Settings
Set the default thinking level in~/.loop/agent/settings.json:
Hide Thinking Blocks
If you want thinking enabled but prefer not to see the blocks:When to Use Thinking
Complex debugging
Complex debugging
Enable
high thinking when the agent needs to reason through complex bugs or multi-step logic.Architecture decisions
Architecture decisions
Use
medium or high when asking for design decisions or trade-off analysis.Simple tasks
Simple tasks
Keep thinking
off for straightforward tasks like file reads or simple edits to save tokens and reduce latency.Model Support
Thinking mode depends on model support. Not all models support chain-of-thought reasoning. When using a model that doesn’t support thinking, the setting is silently ignored.Token usage includes thinking tokens. Higher thinking levels consume more tokens per response.