Interactions API: our primary interface for Gemini models and agents

Source: Google Date: 2026-06-22 URL: https://blog.google/innovation-and-ai/technology/developers-tools/interactions-api-general-availability/

Summary

Google announced general availability of the Interactions API as its unified, stable interface for both Gemini model inference and agent workflows. Key capabilities: server-side state management, background execution for long-running operations, tool combination (Google Search, Maps, image/music generation), and managed agents that provision remote Linux sandboxes. A single API call now handles both synchronous inference and asynchronous agent tasks, with optional background execution for tasks that exceed a single request’s lifespan.

Implications

• MCP / protocol: The Interactions API is Google’s answer to the protocol fragmentation problem — one stable schema for models and agents. This competes with and complements MCP; it will push developers toward Google’s unified surface rather than an open protocol layer. Watch for tension with MCP adoption as the GA API hardens.
• Long-running / autonomous agents: Background execution + server-side state is infrastructure for persistent agents. Google is now shipping the same architectural primitives (durable sessions, remote sandboxes) that OpenAI is describing in Codex-maxxing. Convergence across labs on this pattern is significant.
• Two clocks: Closed-frontier infrastructure move. Managed remote sandboxes and server-side agent state are not replicable with open weights alone — they require hosted infrastructure. Further entrenches the closed/open split at the deployment layer.