Skip to main content

​
Web Search and Reasoning

OpenChat can augment responses with controlled web search and model reasoning traces.
  • Toggle from chat prompt toolbar.
  • Availability is checked per deployment and per-user quota.
  • Daily search limit is enforced in Convex (search.ts).
  • Search tool usage is tracked on message/job metadata.
If search is requested but unavailable, OpenChat records explicit unavailable state and continues with non-search generation.

​
Reasoning

  • Reasoning-enabled models stream reasoning parts separately from answer text.
  • Chain-of-thought style parts are normalized and persisted as structured items.
  • Reasoning metadata includes requested flag, token count, and duration.

​
Stream Metadata

Streaming pipeline tracks:
  • time to first token
  • tokens per second
  • total duration
  • tool call count
  • web search call count
This metadata powers UI analytics and post-hoc performance tuning.