timing
Parameters
timing
When to ingest messages. Defaults to IngestionTiming.ON_LLM_CALL.
IngestionTiming.ON_LLM_CALL: prompt messages are ingested before each LLM call starts, and the assistant output is ingested after completion or stream completion.
IngestionTiming.ON_AGENT_COMPLETION: the final accumulated session prompt/history is ingested once at agent completion.