{
  "id": 17403,
  "title": "Gemini API Managed Agents: 3.6 Flash, hooks, and more",
  "url": "https://urgent.news/2026/07/28/gemini-api-managed-agents-3-6-flash-hooks-and-more",
  "topic": "ai",
  "section": "AI",
  "published": "2026-07-28T16:00:00.000Z",
  "source": {
    "name": "Google Blog",
    "slug": "google-blog",
    "url": "https://blog.google/innovation-and-ai/technology/developers-tools/expanding-managed-agents-gemini-api-3-6-flash-hooks/"
  },
  "original_language": "en",
  "account": "Managed Agents in the Gemini API now offer environment hooks, model selection, and free tier access. These enhancements expand the capabilities of the previously released background tasks and remote MCP server integration. With managed agents, a single API call orchestrates reasoning, code execution, package installation, file management, and web retrieval within an isolated cloud sandbox. Developers can integrate the Interactions API skill by running the command: npx skills add google-gemini/gemini-skills --skill gemini-interactions-api. The Antigravity agent now defaults to Gemini 3.6 Flash, requiring no code modifications. When creating an interaction or managed agent, users can specify a model by passing agent_config.model, allowing them to choose Gemini 3.5 Flash-Lite for lower costs or their preferred model. Environment hooks enable users to execute custom scripts before or after each tool call within the sandbox. By creating a .agents/hooks.json file in the environment, the runtime triggers handlers on pre_tool_execution or post_tool_execution events. Regular expressions in the matcher field allow targeting multiple tools with | or capturing all with *. The hooks documentation provides comprehensive HTTP hook definitions and failure handling guidelines. Companies like AI-native investment bank Offdeal utilize post_tool_execution hooks to automate image verification in the remote sandbox, ensuring company logos meet specific criteria before inclusion in decks. Managed agents are now accessible on free tier projects, enabling developers to experiment with agentic workflows using a project key without active billing. To prevent excessive token consumption, users can set max_total_tokens in agent_config. If the agent reaches the token limit, execution pauses, and the interaction returns \"incomplete.\" The system preserves environment state, allowing users to resume the task by providing previous_interaction_id with a fresh budget. Scheduled triggers automate recurring agent tasks, creating persistent resources that leverage the same sandbox for file persistence. Through the Environments API, users can list, inspect, delete, and recover sandbox sessions or clean them up once the pipeline completes. These updates position managed agents as cost-controlled, autonomous workers that operate within real development environments while maintaining budget control and eliminating the need for external orchestration.",
  "summary": "We’re announcing even more new capabilities in Managed Agents in Gemini API so developers can build reliable, production-ready agents.",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 2,
    "also_reported_by": [
      {
        "outlet": "Google AI Blog",
        "title": "Gemini API Managed Agents: 3.6 Flash, hooks, and more",
        "url": "https://urgent.news/2026/07/28/gemini-api-managed-agents-3-6-flash-hooks-and-more-313298",
        "published": "2026-07-28T16:00:00.000Z"
      }
    ]
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}