Give your long-running agents a sandbox that survives between calls

New in Agno: SuperserveTools, which lets an agent write and run its own code inside a Superserve sandbox. The sandbox is a Firecracker microVM, and the part that matters is that it persists. Files the agent writes and packages it installs are still there on the next tool call, and the next run in the same session. That's the difference between running one-off snippets and actually doing long-running work, where the agent builds something up over many steps instead of starting from an empty box every time.‍

from agno.agent import Agent
from agno.models.openai import OpenAIResponses
from agno.tools.superserve import SuperserveTools

agent = Agent(
    model=OpenAIResponses(id="gpt-5.5"),
    tools=[SuperserveTools(timeout=600)],
    markdown=True,
)

agent.print_response("Fetch the last 30 days of AAPL prices and plot the moving average.")

‍

The secrets handling is worth knowing about. You bind a team secret and the sandbox only ever sees a proxy token. The real credential gets swapped in for outbound calls to the hosts you allow, so agent-written code can hit real APIs without your keys ever landing somewhere the model can read them. You can also switch runtimes with a template, or expose a port to get a public preview URL for whatever the agent builds.

Check out Agno’s Superserve toolkit docs to learn more.