How can I apply filters to a Data Agent in the Conversational Analytics API?

Google Cloud Build with AI Agents
,
1

I’m building with Google Cloud Conversational Analytics (Gemini Data Analytics) and I need a way to restrict or scope the data my Data Agent returns.

What I’m using

from google.cloud import geminidataanalytics

data_agent_client = geminidataanalytics.DataAgentServiceClient()
data_chat_client  = geminidataanalytics.DataChatServiceClient()

I can start a conversation like this:

messages = [
    geminidataanalytics.Message(
        user_message=geminidataanalytics.UserMessage(text=question)
    )
]

conversation_reference = geminidataanalytics.ConversationReference(
    conversation=data_chat_client.conversation_path(
        billing_project, location, conversation_id
    ),
    data_agent_context=geminidataanalytics.DataAgentContext(
        data_agent=data_chat_client.data_agent_path(
            billing_project, location, data_agent_id
        ),
        credentials=looker_credentials
    ),
)

request = geminidataanalytics.ChatRequest(
    parent=f"projects/{billing_project}/locations/{location}",
    messages=messages,
    conversation_reference=conversation_reference,
)

stream = data_chat_client.chat(request=request)
for response in stream:
    print(response)

What I need help with

I’d like to apply filters in a controlled way, either:

  1. When creating/updating the Data Agent so that every query it runs is automatically scoped (for example: always filter by region = 'EMEA'), or

  2. Dynamically per conversation so a user’s chat session can include temporary filters (like a date range) without depending on the LLM to infer them.

My question

Is there an officially supported way—agent-level or per-conversation—to attach filters like these when using the Python client?
Any code snippets, API fields, or best practices would be hugely appreciated!

Thanks in advance for any pointers or sample code

2

Hi @Vamsi_Panchada There isn’t an automatic way for the API to always apply filters, but you can handle it two ways: add a default filter when creating or updating the Data Agent so every query is scoped (like region = ‘EMEA’), or pass a filter in the DataAgentContext of your ConversationReference so it only applies to that conversation. Permanent rules go in the agent config, temporary ones go in the context.

3

Hi @a_aleinikov - i guess there is a no argument in datacontext to apply the filter and if we want to do it through some instructions there is a possibility that it can be bypassed through prompt injection