Transcription
[Music]
[Applause]
Hi, welcome to another video. Today I'll be looking at the Gemini CLI upgrades starting from v 0.7.0 and moving upward. They've launched their extensions feature which I also wanted to talk about. The big idea here is that Gemini CLI isn't just a chat in your terminal anymore. It's a scriptable assistant that can connect to tools you already use, export clean outputs, and streamline workflows, which is pretty cool. It's free to try, the docs are better, and there's a growing ecosystem that feels like a proper platform rather than a single purpose utility.
So, starting with V0.7.0, the headline is the ID plug-in spec. Basically what it does is outline how to build rich context-aware integrations with native in-editor diffing which is quite awesome if you care about review-ready changes in VS Code, Zed, JetBrains, or similar editors. It's very similar to how developer assistants plug into your editor and capture local context, but this spec is designed to feel natural to Gemini CLI flows. If you remember my earlier video, I talked about codeware sessions. This takes that idea and makes it more standardized, which is great for teams who want consistent behavior across editors.
On extensions, there were early versions like Flutter and Nano Banana. Flutter helps you create, build, test, and run apps with Gemini CLI guidance, and Nano Banana is a neat utility integration that generates images right within the AI coder. Telemetry got more flexible via environment variables. So in CI, you can tweak settings without touching JS, which is awesome for pipelines. Experimental todos arrived. You can now turn on a managed checklist for complex tasks, and it gives visible progress tracking that's actually useful. Share chat learned how to include tool calls in the exported markdown. So if you export with `/chat share`, the function calls and responses are captured. Great for PRs and postmortems. Citations are now enabled for all users, which boosts traceability. And custom commands work headlessly now. So you can run something like `gemini /joke Chuck Norris` straight from the shell without opening a full session. I mean, I really liked it. V0.7.0 is a solid quality-of-life release that makes the CLI more composable.
Now, V0.8.0 is the extensions inflection point. Gemini CLI extensions are officially announced with a gallery and simple management. You can install from a GitHub URL or a local folder, then enable, disable, list, update, or scaffold new ones. The idea is simple: connect the CLI to data sources, SaaS APIs, and your own services so the agent can call those tools directly from the terminal, which is quite awesome. There's a new homepage and improved docs on geminacle.com, so getting set up is not bad at all. Non-interactive allowed tools landed, so you can pre-approve trusted operations and headless runs, huge for automation, and terminal title status is a nice touch. Set `show_status_in_title` true, and your window title shows the CLI's live status and thoughts, which is kind of cool when you're juggling multiple panes, plus a bunch of smaller features and polish shipped. You'll notice faster, more reliable interactions and cleaner UIs.
Let's transition into a quick overview. In the terminal, the extensions workflow centers on a few commands. You'll see `gemini extensions install` followed by either a GitHub URL or a local folder path. Then `gemini extensions enable` to turn it on, `disable` to turn it off, `list` to see what's installed, `update` to pull the latest changes, and `new` if you want to scaffold a fresh extension. While you're in a session, you can also use `/extensions list` and `/extensions update` to manage without leaving the flow, which is pretty good.
I'll hover over the extensions gallery in the docs. This is where community, partner, and Google-built integrations show up, ranked by popularity, so you can discover what's there. The homepage links straight to docs and examples. And honestly, the documentation layout is cleaner, which makes onboarding not bad at all. For settings, allowed tools in non-interactive use are the stars if you're automating. In headless mode, the flag is passed at runtime, but you can also define trusted tools and settings so you don't get pestered for approvals. Here's where `show_status_in_title` lives. Flip that to true, and your terminal title updates in real time.
In the IDE integration area, the plug-in spec explains how editors can do native diffing with Gemini's edits, which is great if you're making multi-file changes and want tidy reviews. The share chat feature is under commands. `/chatshare file.md` or `file.js` exports the conversation with function calls included, and telemetry setup is documented with environment variables, so CI and/or level policies are easier to enforce. That's the map: extensions install/enable/update in the terminal, gallery on the site, settings for allowed tools and title status, IDE spec for diffing, and share chat for exports.
The extensions framework makes the CLI feel like a real platform. You bring your stack into the agent, which is awesome for custom workflows. The management commands are straightforward, and the gallery helps you find credible integrations fast. Non-interactive allowed tools are a big win for CI reliability, reducing friction while still staying safe if you choose carefully. The terminal title status sounds cosmetic, but it's genuinely useful context when multitasking. Docs and examples are better, so ramp-up time is shorter.
However, there are trade-offs. Extensions need initial setup. You'll be doing auth configuration and a bit of yak-shaving on the first run, which is a bummer if you wanted instant magic. Headless approvals are powerful, but you need guardrails: sandboxed, clear policies, and a tight set of allowed tools. IDE plug-in maturity will vary by editor. Some will have richer experiences sooner than others. And as with any evolving ecosystem, some extensions will be early-stage, great ideas, but not fully hardened yet.
Personal take: I really liked it and have been using it. That's why I thought I'd talk about this, too. The flow of installing extensions, enabling them, and then just using the CLI as a hub for tools feels natural. It's very similar to how you might wire up LangChain or MCP servers, but with a first-party UX that's integrated and friendly to terminal users, which is really awesome. Claude also did something similar recently. If you're an avid watcher of this channel, you know I care about developer velocity. This set of updates hits that target. Simple commands, better docs, fewer clicks, more control. For teams, the policy and telemetry pieces make it feasible to roll out without chaos. So there's that. I think all this is in turn for Gemini 3, and the terminal interface is really, really good these days. Overall, it's pretty cool.
Now, getting started with building something like this can feel tricky, especially when you hear terms like parameters, mixture of experts, and stuff. That's why I use Brilliant. Their course, "How AI Works," guides you through these concepts step by step, so you can see exactly how these models function. Brilliant makes complex ideas approachable. You start with the basics and gradually take on more advanced problems. They've added plenty of new courses covering foundational math, programming, and AI. And the lessons are designed to be challenging and purposeful, helping you build real understanding. What really sets Brilliant apart is that every course is created by top experts from places like MIT, Google, and Stanford. So, you're learning directly from some of the best minds in the field. You're actively solving problems and experimenting, not just watching. And Brilliant even has an app, which makes it easy to keep learning on your computer or in quick sessions on your phone. You can try Brilliant free at brilliant.org/icodek. And if you go premium, you'll get 20% off an annual subscription, which gives you unlimited daily access to everything on Brilliant. Much love to Brilliant for sponsoring this video and for helping us all learn smarter and build real skills.
Anyway, share your thoughts below and subscribe to the channel. You can also donate via Super Thanks option or join the channel as well and get some perks. I'll see you in the next video. Bye.