Antigravity SDK v0.1.18 adds first-class local model support via LiteRT and LocalOpenAI configurations, letting developers run models on-device or via compatible endpoints. The release introduces a standardized evaluation preset (AgentConfig.eval()) for consistent core coding assessments across agents. Developers can now assign custom subagent models, overriding the root agent’s model on a per-subagent basis. The built-in schedule tool is enabled by default, allowing agents and subagents to create timers and cron-style background jobs.