Hello, I am having a persistent issue with a Dialogflow CX agent that is calling a Data Store Tool. The diagnostic trace shows the tool is being called correctly, but it is not returning any search results from the linked Data Store. To rule out configuration errors, I started over and created a brand new, clean agent from scratch in a new project. I used a simple Flow-based architecture and a basic JSONL data file. The issue is exactly the same, which strongly points to a platform issue. Here are the details of the final test setup: - The agent and Data Store are in the same location (eu multi-region). - The Data Store is confirmed to be indexed successfully (I can see the documents in the console). - The Tool is a Data Store type, with Grounding enabled and set to “Very high”, and uses a strict custom summarization prompt. - Both the Dialogflow and Discovery Engine APIs are enabled. - The test fails with a simple, English-only JSONL file and a direct English question. The final agent response is “I don’t understand.” Has anyone encountered a similar backend issue where a correctly configured Data Store Tool consistently fails to retrieve results? Any help would be greatly appreciated.
Hi Nikita_Lopatenko,
It looks like you are facing a persistent issue where your Dialogflow CX Data Store Tool is being triggered correctly but consistently fails to retrieve any results from your linked Data Store. Even after you’ve double-checked your configuration and rebuilt the setup from scratch, the tool still doesn’t return the expected data, which leads to fallback responses like “I don’t understand”, a clear sign that your system isn’t successfully retrieving the information it’s supposed to.
You might find it helpful to check this page, it discusses a missing API feature that prevents customizing summarization prompts for Dialogflow CX Data Store Tools, which could be impacting your agent’s ability to generate accurate or relevant responses.
In addition, consider simplifying or temporarily removing your custom summarization prompt, experimenting with different grounding levels, double-checking that your JSONL schema is correctly structured and contains clear, relevant content, reviewing the Discovery Engine API request and response in the diagnostic trace for any anomalies, and testing with an alternate Data Store that’s known to function properly to help isolate the issue.