Skip to main content
When a reasoning model is running, Foundry shows a Reasoning block above the assistant’s reply that streams the model’s chain of thought in real time. Click the block to expand and watch. Collapse it once the reply lands — the trace stays attached for later review.
Screenshot coming. Expanded reasoning block, mid-stream.

​
What you’ll see

  • Brain header with an animated spinner while thinking.
  • A monospace stream of the reasoning trace.
  • The final answer below, in normal chat formatting.
If you collapse the block while it’s streaming, Foundry keeps streaming in the background; expand again any time to see where it’s at.

​
Why it matters

  • Debug bad answers by reading what the model considered.
  • Build trust — when the model is confident, the trace tends to be linear; when it’s uncertain, you’ll see it backtrack.
  • Learn faster — the trace often surfaces angles you hadn’t thought of.

​
Tips

  • The trace can be long. Collapse and skim the answer first; expand the trace only when you want to dig in.
  • For repeated debugging, set Reasoning Effort to High to get the most-detailed trace.

​
Limits

  • The trace is not the model’s reasoning verbatim in the philosophical sense — it’s the model’s externalized scratchpad. Useful, but not a brain scan.
  • Some providers redact parts of the trace; you’ll see those as [redacted] markers.

Effort control

Pick how deep the trace goes.

Reasoning models

Concept overview.