Sign InOpen Brain
GoogleOfficial ReleaseOfficial Source

Gemini API Managed Agents: 3.6 Flash, hooks, and more

Gemini Managed Agents now add model selection, sandbox hooks, token caps, schedules, and free-tier access, covering several controls needed for unattended coding work.

Google · Jul 28, 2026
Open Source Open MarkdownOpen JSON
Source Summary

Gemini Managed Agents now default to **Gemini 3.6 Flash**. The service adds pre- and post-tool hooks, explicit model selection, scheduled triggers, free-tier access, environment management, and persistent files across scheduled runs.

Practical Implication

Add hooks where tool calls need policy checks, linting, or audit output, and set **max_total_tokens** on autonomous loops. A capped run pauses as incomplete while preserving its environment, so it can resume with a new budget and previous interaction ID.

Agent-Ready Context
Gemini Managed Agents now default to **Gemini 3.6 Flash**. The service adds pre- and post-tool hooks, explicit model selection, scheduled triggers, free-tier access, environment management, and persistent files across scheduled runs.

Add hooks where tool calls need policy checks, linting, or audit output, and set **max_total_tokens** on autonomous loops. A capped run pauses as incomplete while preserving its environment, so it can resume with a new budget and previous interaction ID.

The agent remains a preview identified as antigravity-preview-05-2026. Scheduled runs reuse a sandbox, which is useful for stateful work but makes cleanup, state drift, and hook failure behavior important operational concerns.
Connected Context · Feed7 Judgment

This advances Gemini Managed Agents from asynchronous tool execution toward an operable autonomous service: hooks add policy boundaries, schedules and persistent files enable recurring stateful work, and token caps bound loops without discarding state. It also concentrates new failure modes around reused environments, drift, cleanup, and hook behavior, while preview status limits production confidence.

Expanding Managed Agents in Gemini API: background tasks, remote MCP and moreThe earlier update established asynchronous execution, remote MCP, client functions, and persistent sandbox state; this one adds scheduling, hooks, explicit budgets, and environment management around that foundation.From fork() to Fleet: Designing an Agent Sandbox Cloud — Abhishek Bhardwaj, OpenAIPersistent scheduled environments align with the sandbox-cloud case for isolated, recoverable state, while making cleanup and drift concerns concrete at the managed-agent API level.Project-scoped TokensPre-tool hooks can enforce policy before execution, while project-scoped tokens constrain what execution can reach; together they represent complementary procedural and credential boundaries.Vercel MCP can now deploy codeDeployment is the kind of consequential tool call that benefits from pre-call approval checks and post-call audit output, though hooks do not by themselves supply rollback or prove deployment success.
Context Map
toolscodingsecurity#cloud-agents#sandboxing#tool-use
Uncertainty
The agent remains a preview identified as antigravity-preview-05-2026. Scheduled runs reuse a sandbox, which is useful for stateful work but makes cleanup, state drift, and hook failure behavior important operational concerns.