Crynta’s work is a masterclass in architectural minimalism, proving that powerful development tools don't need to be bloated. It is a refreshing reminder that elegant engineering can still achieve more with less in an era of resource-heavy software.
Approfondir
Prérequis
- Pas de données disponibles.
Prochaines étapes
- Pas de données disponibles.
Approfondir
My 7MB AI Terminal Is Now a 8MB Full Dev WorkspaceAjouté :
2 weeks ago, I made a video about turks.
Some of you watched, some of you just started. And even better stack made a video about it. Thank you so much for all the support and the feedback. Since then, I've been actively working on it and here is what changed.
A lot of you asked for more customization. You can now pick from built-in themes, use custom background images, or even drop in a GIF if that's your thing.
And you can create your own theme from here. It will open see the editor, so you can set your colors, save the file, and you are done. You can easily share it with others as well.
Anyway, I will keep improving the customization layer a lot going forward.
So, just say what you actually need here.
Next step is the source control. I wanted a proper Git workflow built directly into the interface. So, now, right from the sidebar, you can stage your files, check your changes, and write your commits. And if you don't want to write the commit message yourself, the AI can do it for you. It actually reads all the diffs, so the message make perfect sense every time.
There is also Git graph tab now. You get your full commit history at a glance.
Click any commit to see who made it and what changed. If you click a file, the diff opens right in the editor, so you don't have to jump between apps to review your code.
Terminal splits have landed, too. You can now create horizontal and vertical panes side by side, either from the UI or with fully customizable keyboard shortcuts. And the file explorer is context aware. If you have two different projects open in splits, the sidebar will automatically show the project for the pane I'm currently focused on. It makes the whole experience feel like one unified workspace. You can also create a file and it will instantly be available from the file explorer.
I also added a smart notification system for coding agents with first-class support for our Tarox AI agent and cloud code.
Codex and Gemini CLI coming soon.
So, here, if an agent needs your permission for an action and you are in another tab, you get clean in-app notification. If you switch to completely different application, it fires a native OS notification, so you never lose your context.
To make this even better, there is a new cloud code command. It analyzes your requests, balance a fresh cloud code instance, hands it a detailed task, then verifies the work when it's done.
This kind of a agent orchestration is really interesting aspect for Teraclassing.
Moving forward, the goal is to control multiple agents in parallel, distribute a specific task for each, review their outputs, and cleanly merge the result.
Or maybe something even better. Let me know if you have some specific ideas for this.
By the way, if you're on Windows, WSL now works out of the box.
On the AI side, the entire provider system has been redesigned. Alongside the existing providers, support has been added for many other providers, including Open Router and Deepseek, as well as local models via Llama and MLX.
To attach a file from your project, just type @ in the chat, search for the file, and it's instantly available to the agent.
Because Tarox now runs across all major desktop platforms, such as macOS, Linux, and Windows.
No, not mobile yet.
I'm natively testing and optimizing across different desktop environments and shells before every release, which turns out to be a lot harder than it sounds.
And speaking of optimization, keeping the app lightweight and fast takes constant work. Every step matters.
Just one small example. We used to universally AI's the code. It's great for UX, but keeping things light for wasn't exactly a priority. By default, it uses Shiki for syntax highlighting, which relies on heavy web assembly bindings.
But since CodeMirror was already powering the main editor, having two separate syntax highlighting engines in the same app just didn't make much sense. So, I rewrote this part, ripped out Shiki, and wired the eye chart to use our existing code mirror engine.
It's actually faster and drops a lot of unnecessary bundle size.
So, yes, using Termux correctly is a big part of it, but it's more architecture decisions like this that keep the app not just at 8 MB, but incredibly fast. I will do a full breakdown of this in the next technical video.
Currently, I use Termux as my daily driver, even though it's not 1.0 version yet, it's already better to me than any other tool I've used.
I just wanted one fast, lightweight tool with an actually great UI and UX, something that covers all the real-world development needs of modern engineering while staying under only 8 MB.
But, it's not just about me. I really want to make Termux the ultimate tool for everyone else, too.
I want it to work exactly the way the community needs it to work, driven entirely by your feedback.
Termux is completely free, and it will always remain open source.
So, here is what I'm working on next.
First, real-time auto complete right in the terminal.
Then, pluggable LSP. This will be optional, so if you want the full IDE features, you will be able to download it separately, keeping the core app completely lightweight.
I'm also bringing in persistent terminal sessions and layout restoring inspired by tmux with vertical tabs.
Plus, native SSH and remote control support, a better web preview, even more performance optimizations, and a lot of other things.
I really appreciate every single pull request, bug report, idea, and piece of feedback you guys have dropped in the repo over these 2 weeks.
To make all of this better, I created a Discord server for the project. Termux has zero telemetry, so Discord is the best place to share ideas or let me know if something breaks. Links are in the description.
Oh, and by the way, I just open sourced Taras' website, too. And in general, I have some interesting and complex private projects which probably I will polish and open source them in the future, as well. Thank you again for being part of this. More coming soon.
So, see you in the next video.
Vidéos Similaires
OpenHuman VS Hermes AI: Who Wins?
JulianGoldieSEO
285 views•2026-05-29
BREAKING: Microsoft’s New Image Generating Model Beat Out GPT 1.5 and Nano Banana 2
aimmediahouse
122 views•2026-06-03
Long-Running Agents — Build an Agent That Never Forgets with Google ADK
suryakunju
142 views•2026-05-30
This computer is made from real human brain cells. And you can buy it.
Talktmsmedia
3K views•2026-05-28
I Made the Same Anime Fight Scene in Every AI Video Generator
NobleGooseAnime
295 views•2026-05-30
Nvidia Bets Big On AI PCs | New Chip To Power Windows Laptops | Technology | AI Updates | N18S
cnnnews18
3K views•2026-06-01
I Tested NEW Opus 4.8 on Four Projects (Updated LLM Leaderboard)
AICodingDaily
298 views•2026-05-29
3D Platformer Update - NO CAPES
SolarLune
294 views•2026-05-30











