Managed Deep Research and Antigravity

Managed Deep Research and Antigravity

You can now run Google's two most capable managed agents, autonomous research and a code-running sandbox, without leaving the Gemini setup you already have.

GeminiInteractions reaches both of them through the same model class. You switch one on by setting the managed agent in place of a model ID, so there's no new client and no separate integration to wire up.

Agent 1: hand off the research, get back a cited report. Deep Research plans the task, browses the web, and returns a researched report with citations. It runs in the background, with streaming you can reconnect and resume after a dropped connection, so a long job survives a flaky network instead of starting over.

from agno.agent import Agent

from agno.models.google import GeminiInteractions

research = Agent(

model=GeminiInteractions(id="deep-research-preview-04-2026"),

)

research.print_response("Research solid-state battery commercialization.")

‍

Agent 2: hand off the work, get back the artifacts. Antigravity is a general-purpose agent that plans, runs code, and produces artifacts inside a managed Linux sandbox, so it does the work rather than just describing it.‍

sandbox = Agent(

model=GeminiInteractions(id="antigravity-preview-05-2026"),

)

sandbox.print_response("Summarize the top 5 Hacker News stories as Markdown.")

‍

Same model class, same interface you already use for Gemini. Swap the ID for a managed agent and you're running.

Managed-agent setup, including the config knobs for each, is in the Agno docs: https://docs.agno.com/models/providers/native/google/gemini-interactions