I made a blog! And also improved how the skills look by using icons instead of text.

Devlog #3

I gave the agent a browser and the ability to control it so now it can test the frontend to whatever it makes to further shorten the gap between this and paid vibecoding tools. I also added some new small features to make some dev workflows easier, and I made diff reviews way better and made undoing changes not require a github repo.

Devlog #2

I just completely revamped the MCP settings. It went from just a JSON editor to a new UI that supports fancy auth, like the ChatGPT implementation.

Devlog #1

I just made real vibecoding free.

I created Opencodex, a VS Code agent extension that gives you a workflow almost identical to OpenAI Codex, but without requiring a paid AI subscription or actually, any account at all to start. It lets you use free models from 5 different cloud providers, as well as Ollama for local models.

In just one week on the VS Code Marketplace, Opencodex has already passed 480 installs. The main difference from competitors like Cline, Continue, and Kilo Code is that while those tools are free to install, you generally still need to bring your own API key or pay for model usage. Opencodex focuses on making the entire experience accessible for free. I’m working on making MCP more intuitive by removing the current bland JSON editor to a ChatGPT style MCP manager.

You can install it here: https://marketplace.visualstudio.com/items?itemName=EvaanChowdhry.opencodex-agent

I just fixed macOS compatability (to some extent). Because of it’s nature I had to resourt to an alternative to PTY style terminals. I also fixed the settings menu (stale accent color picker) and added new features such as manual update checks, and syncing the terminal path to the file explorer.

I’ve been debugging this theme editor for the past ~20 minutes. It turns out that VSCode live preview is NOT as trustworthy as you think it is. I’ve been testing using live preview and it works perfectly fine. But as soon as I push into production, there’s 404s in the console. So I’ve been trying to fix them, I’ve redeployed too many times but I just fixed it. I also updated my prod.sh to completely handle updating for me. It creates new commits, releases, etc etc based on what kind of update I did. So a site update won’t trigger a new release now like it did before.

I just added a theme editor! This is linked on the main website (termiview.xyz) and it lets you design custom termiview themes so it doesn’t look as bland as it does right now. Termiview instances now have a capability to load these custom themes ofc

Termiview is essentially an ssh client without the ssh designed to make navigating your server easy. It has a file explorer that lets you view, edit, and move your files with a GUI instead of via commands. It also has a terminal so you can actually do stuff on your server. I plan on later adding an AI mode to this which can either use local or cloud models to execute tasks on your server, so basically set things up FOR you when you tell it to. Oh and it also works on macOS. This first devlog gets the basis ready, the site, and the actual termiview instances itself. It’s perfectly usable at this point in time

I spent some time working on 4 new commands. There’s also custom modes now! Each slack channel can have its own custom mode. Same way how each channel can have its own game. I also added a new /end command. If the person who started the game runs it, the game in the current channel will end so others can be played!

I got the first 3 commands working! So basically this is a slack bot based on my old project Numero for Discord. It shares a similar idea of being a number guessing game. So far I’ve managed to get the /start /guess /hint working and a full game system! I’ll keep working on it for the next few hours to try and get more of the key commands in. Maybe a few more things that weren’t in the discord bot like custom gamemodes, we’ll see!

So I originally planned to release a new blog feature which would scrape my Linkedin profile for each post so I don’t actually have to write anything. But it turns out that the way to do that would’ve been very tedious since Linkedin doesn’t have an API for this type of stuff. So instead I decided to add a project status tracker to each project I have up right now since one of my older domains just expired. It reminded me that these ones might too and I might not remember them.

So this is my new portfolio site! I made it a single page, designed in a modular layout so it’s way easier to change things when need be. It also has a really cool feature where it shows the song I’m currently listening to on Spotify near the contact buttons, but that media player only shows when I actually am listening to something. It features a clock telling users the current time in my timezone, and all appropriate contact links. This is really different from what my portfolio used to be: a macOS terminal lookalike with commands for each one of these current cards.

Well, I just had to migrate Peylo’s supabase account since I forgot the email to the last one. Now instead of using Google Auth, it just uses Supabase auth links. This also means the 2 full units I had generated in the last DB are now gone, so I have around 12 lessons now for the first unit. More to come throughout the dev process. Now at least it’s fully functional.

I just brought Peylo back to life, its a project I created a few weeks ago in June, I had originally given up on it due to Supabase disabling the project because of inactivity. But after I saw Hackclub Stardance, I decided to give it a shot again. Peylo is a learning platform designed to mimic Duolingo’s ease of use but instead of learning languages you learn financial literacy. Now you might think: “How do you make each lesson?” as there’s around 500 lessons in Peylo’s curriculum. Well, I created a Gemini powered lesson generator to automatically generate each lesson and store it in the DB for future use. Now, you might think how I’m managing my Gemini API usage; the truth is, I don’t. I created a system with around 6 different Gemini API keys from all 6 of my Google accounts, and it automatically switches to the next one after one hits its limit. I’m looking forward to continuing on this project in the next few months.