Core Updates Overview
Google’s Gemini CLI v0.40.0, released April 30, brings four substantial upgrades, all pointing in one direction: making the AI Agent in the terminal more like a true collaborator, not a one-time Q&A machine.
| Feature | Description | Significance |
|---|---|---|
| Tiered Memory System | Cross-session memory + project-level context | Agent starts “remembering” your project |
| Auto Skill Generation | Automatically creates reusable skills from session history | From “repeated teaching” to “automatic learning” |
| Gemma Local Support | Can run Gemma models locally | Zero cost for simple tasks |
| Streamlined UI | Compact toolbar and topic view | Improved terminal space utilization |
Tiered Memory System: Agent Finally “Gets a Brain”
Before v0.40.0, every Gemini CLI launch was a fresh conversation. The tiered memory system solves this with three levels:
┌─────────────────────────────────────┐
│ Global Memory │
│ User preferences, coding style │
├─────────────────────────────────────┤
│ Project Memory │
│ Architecture decisions, tech stack │
├─────────────────────────────────────┤
│ Session Memory │
│ Current dialog context, temp vars │
└─────────────────────────────────────┘Auto Skill Generation
The most underrated feature of v0.40.0:
- You teach the Agent a task
- Agent automatically analyzes the pattern
- Automatically generates a reusable skill
- Next time, automatically applies it
Cost Breakdown with Gemma Local Support
| Task Type | Routing Target | Cost | Latency |
|---|---|---|---|
| Simple queries | Local Gemma | $0 | < 500ms |
| Medium complexity | Cloud Gemini Flash | Low | ~1s |
| Complex reasoning | Cloud Gemini Pro | Medium | ~3s |
Getting Started
- Upgrade CLI:
npm update -g @google/gemini-cli - Enable tiered memory (default on)
- Auto skill generation: Prompted on first use
- Local Gemma config: Set in config.json
Gemini CLI v0.40.0 is not a “minor update” — it’s Google’s paradigm statement in the terminal Agent space.