Gemini API agents get cron scheduling and free-tier access
Managed Agents also lands the governance builders wanted: pre/post tool hooks, hard token-budget caps, an Environments API, and 3.6 Flash as the default.
Copy markdown
Cron for agents, no orchestrator
Bind a managed agent to a cron schedule and it runs on its own inside a persistent sandbox where files survive between runs. A nightly-triage or weekly-report agent no longer needs a separate Cloud Function or scheduler bolted on the side.
Hooks that can veto a tool call
A new .agents/hooks.json runs your scripts before and after every tool call; return {"decision":"deny"} from a pre-hook and the call is skipped, while post-hooks format or audit the result. Regex matchers scope them to one tool or all of them — real guardrails without wrapping the whole loop yourself.
A token cap that actually stops
The new max_total_tokens parameter caps input, output and thinking tokens for a run; hit the ceiling and the agent pauses with status "incomplete", saving state instead of looping into a surprise bill. It is the runaway-cost backstop agents have been missing.
Gemini 3.6 Flash is the default now
Agents default to 3.6 Flash at the same $1.50/$7.50 per Mtok as 3.5, but with ~17% fewer output tokens on comparable tasks and a 12-point DeepSWE jump (37% to 49%). Switching is just the model ID; drop to 3.5 Flash-Lite if you want cheaper.
Free-tier keys can run agents tonight
Managed agents now work on free-tier projects with non-billing API keys, and a new Environments API lets you list, inspect and delete sandbox sessions instead of waiting out the 7-day TTL. You can prototype a scheduled agent for $0 before turning on billing.