Currently the only way to find AI conversations is get_ai_conversation_details, which requires a known gen_ai.conversation.id. There is no tool to discover or list conversations — the Explorer Agent has to cobble together a search by querying the spans dataset manually (e.g. span.op:gen_ai.* has:gen_ai.conversation.id) and then iterate individually.
- Add a conversation-search capability to the
search_events flow (spans dataset) so agents can find conversations by time range, user, project, or other span attributes without needing a pre-known conversation ID
- Likely approach: expose a dedicated
search_ai_conversations tool or add first-class conversation-search guidance and field hints to the existing search_events agent config for the spans dataset
- Unblocks use cases like "fetch interesting conversations from project X in the last 24h" that currently require multi-step workarounds and return no results when user fields (e.g.
user.email) are not instrumented
Action taken on behalf of David Cramer.
Currently the only way to find AI conversations is
get_ai_conversation_details, which requires a knowngen_ai.conversation.id. There is no tool to discover or list conversations — the Explorer Agent has to cobble together a search by querying the spans dataset manually (e.g.span.op:gen_ai.* has:gen_ai.conversation.id) and then iterate individually.search_eventsflow (spans dataset) so agents can find conversations by time range, user, project, or other span attributes without needing a pre-known conversation IDsearch_ai_conversationstool or add first-class conversation-search guidance and field hints to the existingsearch_eventsagent config for the spans datasetuser.email) are not instrumentedAction taken on behalf of David Cramer.