I’m using the Agent Designer to build a no-code agent in Gemini Enterprise (formerly Agentspace).
I want to make programmatic calls to one of the agents I created.
I can seen the agent when I use the list all agents call:
curl -X GET \
-H "Authorization: Bearer $(gcloud auth print-access-token)" \
-H "Content-Type: application/json" \
-H "x-goog-user-project: $GOOGLE_CLOUD_PROJECT" \
"https://discoveryengine.googleapis.com/v1alpha/projects/$GOOGLE_CLOUD_PROJECT/locations/global/collections/default_collection/engines/$ENGINE_ID/assistants/default_assistant/agents" | jq
While there are ways to query Deep Research with an HTTP call to the API, I can’t query the no-code agent, I’ve tried multiple different permutations of assistants, default_assistants, agents, IDs and fully qualified names in the URL and the payload, but no luck.
The no-code agents for Gemini Enterprise are in preview mode so it wouldn’t be very surprising if they are just not exposed at the API level, and this is also implied in the docs.
However, in my experience, the no-code agent has a better control of context and instructions than the “Additional LLM system instructions” alone in normal GE calls.
I really want to be able to utilize the ease of Data Stores and the overall RAG experience of GE + no-code agents, but this inflexibility may require me to write a custom ADK solution, maybe leveraging the RAG sample.
Any suggestions?