Dynamic filter when using VertexAiSearchTool with bypass_multi_tools_limit=True #4597

When bypass_multi_tools_limit=True is enabled, the underlying implementation switches over to DiscoveryEngineSearchTool, which doesn't expose the same _build_vertex_ai_search_config hook.

To achieve a dynamic, state-based filter with DiscoveryEngineSearchTool, you can subclass it and override its configuration builder method (or customize how it handles the search request parameters at runtime using the ReadonlyContext).

Here is an example approach you can use:

Python
from google.adk.tools import DiscoveryEngineSearchTool
from google.adk.types import ReadonlyContext

class DynamicDiscoveryEngineSearchTool(DiscoveryEngineSearchTool):
def _build_discovery_engine_config(self, ctx: ReadonlyContext):
# Extract the user_id dynamically from the context state
user_id = ctx.state.get('user_id')

    # Get the default configuration
    config = super()._build_discovery_engine_config(ctx)
    
    # Inject or update your dynamic filter
    config.filter = f"user_id = '{user_id}'"
    return config

Alternatively, if your version of the tool builds the search request dynamically per tool execution, you can also wrap the tool call to inject the context state right before execution.

Hope this helps point you in the right direction!