Deepmind agents run offline now

google deepmind introduced four features for gemini api managed agents: background execution for asynchronous tasks, support for remote model context protocol (mcp) servers, custom functions alongside built-in tools, and credential refreshes that preserve sandbox state. these upgrades allow agents to interact with internal databases and operate without persistent client connections, streamlining enterprise integration. available via the gemini interactions api, this moves towards more autonomous, persistent infrastructure for ai agents.