Gemini API Managed Agents Now Default to 3.6 Flash with Hooks
Google DeepMind has introduced an update to its Gemini API Managed Agents, making the Gemini 3.6 Flash model the standard option. Developers can now use new environment hooks to run custom scripts before or after tool calls in the sandbox, which helps with processes like blocking, linting, and auditing. The update includes budget controls via max_total_tokens, scheduled triggers with cron, and a free tier for testing. The available models consist of Gemini 3.6 Flash (default), 3.5 Flash, and 3.5 Flash-Lite. Users can manage sessions with the Environments API, and OffDeal, an AI-driven investment bank, uses post_tool_execution hooks for automatic image checks. Managed agents can handle reasoning, code execution, package installations, file management, and web retrieval through one API call, all within a secure cloud sandbox. The antigravity-preview-05-2026 agent now defaults to Gemini 3.6 Flash without requiring code changes. Users can select models via agent_config.model, and hooks can support HTTP handlers that POST to external endpoints. Scheduled triggers integrate an agent, environment, prompt, and cron schedule into a lasting resource. Free tier projects can access managed agents using an API key without the need for active billing.
Key facts
- Managed Agents now default to Gemini 3.6 Flash.
- New environment hooks allow custom scripts before/after tool calls.
- Budget control via max_total_tokens caps total consumption.
- Scheduled triggers enable recurring agent tasks with cron.
- Free tier access available for experimentation.
- Supported models: 3.6 Flash, 3.5 Flash, 3.5 Flash-Lite.
- Environments API for listing, inspecting, deleting sandbox sessions.
- OffDeal uses post_tool_execution hooks for logo verification.
Entities
Institutions
- Google DeepMind
- OffDeal