Gemini CLI v0.40: Tiered Memory System, Auto-Generated Skills, and Local Gemma Model Routing

GitHub / Gemini CLI·medium signal

Google's Gemini CLI shipped v0.40.0 with three significant features: a tiered memory system replacing the legacy MemoryManagerAgent, automatic skill generation from past sessions, and experimental local Gemma model routing that keeps routing decisions on-device instead of sending them to a hosted model. The local routing feature is notable — it means routing decisions (which model handles which task) can happen offline, reducing latency and API costs. Direct competitor evolution to Claude Code's workflow.