CodeInOven

The way I've been using CodeInOven lately has been making me rethink the layout of the whole app. I created an assistant that tracks our products, reports about it's online presence, SEO, Google search console, bing webmaster, cloudflare and posthog analytics. I got a report about a 2 star rating on Milogs, then I asked the assistant to inspect it on the same thread, it found it and confirmed the issue. Then asked it to get the spec of the reviewer's decide from play console then setup an android emulator that has lower specs to pressure test the app, it did that, all from the assistant thread. Meanwhile I have this repo opened as a project on the project view, but I've been doing more of my work from the assistant view, connecting marketing, customer support, error tracking all from one assistant. I think the idea of thinking of a product as just a coding project is the wrong mentality. We should start thinking vertically and use our agentic tools with that mentality, the result will surely be astronomically better

Yes, when your agentic app is doing work, the thing under it is usually grouped with it, and electron is blamed for everything, but how about you don't open your browser while using your agentic platform? How do I check my work? Use the same agentic app, most of them come with it now. The in app browser is not as good? Use the one in this thread and see if it fits. Use an agentic app that makes you depend less on other apps so you system can breathe. Or connect to a cloud VM or another machine and work with that

> open twitter > See a new open weight model drop > Check model's size, see anything greater than 1gb > Close hugging face > Come back to twitter and look for another gem.

The whole apple store experience as a developer is trash. You delisted my app, I was not notified, it has been delisted apparently for almost a month, i didn't know because I don't use ios, only found out after prospects started saying it's not on iOS. I contacted you multiple times, no response. Submitted a new version, no response Tried to reach out via twitter, dm is unavailable. Clicked the link on their support page, and it's taking me back to support page where I was not responded to in the first place. No problem.

My least favourite thing to do. My entire mac is practically a toaster now, and this has been going on for hours. It is what it is. We are currently training flotxt for milogs on top of needle 3, an upgrade from needle version ~1 So in the next version of Milogs, expect: - A new speech model (whistle by the cactus team) - A smarter milogs-transformer for your autonomous actions - Semantic search powered by google's EmbeddingGemma 2 And other optimizations and bug fixes. Your app should feel like butter when we release version 1.0.148

We actually cooked with this.

Here is to pass a secret to your agent safely. Works for anything. API keys, emails, passwords, PIID, etc

Rick rolling is baked into the models. I asked deepseek to do a test on YouTube links not playing in a new view. The first video is the one I pasted and the second deepseek. I got Rick rolled hard. I was genuinely confused at some point 🤣🤣

I've never understood why memory is considered a hot topic. Maybe it's because of the context in which I use it, and the controversy is on other context like assistant work. Well I don't have enough exposure in that regard to comment, but I have tremendous exposure on the coding side and I can tell you categorically. Memories should be proposed, not auto saved. We should generally be building AI products with the intention of a person signing off on it How it was built in @codeinoven is two ways: 1. Proposal 2. Definition - Proposal An independent model evaluates the message sent to the agent, then decides to propose a memory based on various criteria. Does this sound like a one off feature, a vent or something that should be remembered. It takes the previous user's message if available as an additional context. It then proposed based on category, priority and scope. Category involves behaviour/preferences/etc; scope is global/project/thread Then it is sent to a memory bank for approval and the user has the option to see the proposed memory and approve, it is discarded after 7 days. - Definition If you want the model to remember something, you define it yourself, using the same rules proposal uses (category/scope/priority) This works all the time and it has improved my experience working on projects significantly. I believe we can use this same principle in other sectors of AI products too. The moment we start dropping the notion of assuming for users, things like this will not even be up for discussion

Ever since I started working on @codeinoven the moment I start working on another project/client project. I get distracted by the fact that. It would be nice if I could simply do this and that. The latest distraction was putting a global browser where I can open x to see updates. I understand this is the woman in the red dress dilemma. I have even decided that today I will not work on any features, yet, here I am, thinking for making it possible to simply copy a huggingface link, and cio will check the model against your hardware to confirm if it can run, download it, set it up you then make it available to you via pi 🤦🏾‍♂️