Deep search

Deep search lets you find relevant traces using natural language queries on the Logs page. Unlike keyword search, which looks for exact text matches, deep search uses AI to evaluate each trace's semantic relevance to your query. Results from your deep search query stream live into the Loop panel.

Some examples of deep search queries include:

• "Show me examples where users are struggling"
• "Find conversations with frustrated users"
• "Highlight cases where customers are happy"

To enable deep search, navigate to Settings > Feature flags and toggle on Deep search. Deep search uses GPT-5-mini and GPT-5-nano, so make sure you have OpenAI configured under Settings > AI Providers.

On the Logs page for a project, click Loop to open the Loop panel. Click Deep search to enable deep search mode. Enter a natural language query (e.g., "Show me examples where users are struggling") and press Enter. Up to 20 traces matching the query will stream into Loop. Select Apply as table filter to apply the query as a table filter. The filter only includes the traces that were sampled and evaluated during deep search. You can then save the view with the deep search filter applied.

See also

• Loop
• Views