Gemini API managed agents default to 3.6 Flash with environment hooks
This update formalizes safer, cost-aware autonomous workflows and widens access for experimentation by enabling new safety and governance features in managed agents.
At a glance
- Gemini 3.6 Flash is now the default for Gemini API Managed Agents.
- Environment hooks let you block, lint, or audit tool calls inside the sandbox via a configurable setup (e.g., .agents/hooks.json).
- Budget controls, scheduled triggers, and free tier access are now available for managed agents.
- To cap resource use, max_total_tokens can be set in agent_config; when reached, execution pauses with status “incomplete” and the state can be resumed later.
The story
Google’s blog post on Gemini API Managed Agents announces that Gemini 3.6 Flash is now the default for managed agents, building on prior updates that introduced background tasks and remote server integrations.
A key upgrade is the introduction of environment hooks, which allow developers to run pre-tool and post-tool execution scripts inside the agent’s sandbox. The documentation notes that you can define hooks in a file like .agents/hooks.json and tailor actions for specific tools or events.
The post also highlights new governance features such as budget controls and scheduled triggers, along with free tier access for projects without active billing, making it easier to experiment with agentic workflows.
Crucially, the system supports token budgeting via max_total_tokens in agent_config. If the limit is hit, execution pauses and returns an incomplete status, preserving the environment so you can resume with a new budget and a previous_interaction_id. The update also reinforces the ability to explicitly select models through agent_config.model, including options like gemini-3.5-flash-lite.