Skip to main content
Central config for the reasoning layer.

​
API key

Your Anthropic key, encrypted at rest per-user. See Bringing your API key.

​
Model selection

Two choices for research + agentic paths:
  • Sonnet 4.6 — recommended default. Fast, capable, ~5× cheaper than Opus.
  • Opus 4.8 — most capable. Use for hardest reasoning tasks.
  • Haiku 4.5 — fastest, cheapest. Good for lightweight jobs.
Change applies immediately to new runs; in-flight runs finish on their original model.

​
Research window

Default 90 days. See Research window.

​
Cost visibility

Every LLM call is logged in Activity log (Settings tab) with token counts. Use it to see where your spend is going.