Skip to content

Execution mode and result visibility

Control how tools run and what the language model sees after execution — without changing your API.

ModeBehaviorBest for
syncChat waits for the tool to finish before continuingFast lookups (< few seconds)
asyncTool runs in the background; visitor gets updates via status notificationsSlow APIs, batch jobs, multi-step backends
SettingModel receivesBest for
llmFull tool JSON (subject to size limits)Default — answers need API data
noneSuccess/failure only, no payloadSide-effect tools (create ticket, send email) where the model should summarize generically

Use none when responses contain PII you do not want in the model context, or when the assistant should only confirm action completion.

ScenarioExecutionVisibility
Order status lookupsyncllm
Hotel offer search + carouselsyncllm (+ result_transform)
CRM note creationsyncnone
Report generationasyncllm
  1. Open tool → Execution.
  2. Set Execution mode and Result visibility.
  3. Pair async mode with on_start / on_delay notifications for better UX.

Rich content cards are built server-side from successful tool JSON regardless of visibility — visitors see cards in the widget; none only limits what the LLM reads for follow-up turns.