CookieAI

DEVLOG #6 – Video implementation… I give up
I genuinely don’t get it anymore.
I tried everything. I read docs. I experimented.
I even asked Gemini how to integrate video… and somehow it didn’t know either

This has been a massive headache, it’s clearly not production-ready, and I doubt it’ll be used enough to justify the pain so yeah, I’m dropping it for now before it drops me.

Next up:
Bringing OpenAI back
Building a proper web settings/config page (finally)
Making the whole setup way more intuitive and less “pray it works”

For now, I’ll focus on my other project… or I’ll just spawn a new one out of pure exhaustion.
http://159.65.96.245 BEST HELPING AI HERE!
See you soon :3

DEVLOG #05 – Better CookieAI Design

I massively improved the UI.
Removed the provider, model selection, and clear chat buttons.
Coming soon: an Instant / Thinking toggle so you can choose how you want responses.
I also improved the image system and I’m currently adding video support for Gemini.
These videos of devlogs take a lot of time to make, so please support the project 💙

See you soon :3
http://159.65.96.245

DEVLOG #04 – CookieAI

Refactored the code.
Dropped OpenAI and Anthropic support — Gemini is enough for now.
Keeping everything compatible was getting stressful, so simplifying helped a lot.
I’m currently working on Gemini’s multimodal (video) system.

Optimizations
Cleaned up API responses (less chaos, more stability)
Adjusted rate limits so the backend doesn’t kill itself
Better handling of conversation history and large images
Fewer random errors, more “it just works”
Better animations

Key takeaway
The backend doesn’t need to be funny, it just needs to not break.
Everything is calm for now.
For now.
The top website for the top.
See you soon :,)

http://159.65.96.245

DEVLOG #3 - CookieAI now can see!
#ImageHell Almost Nuked the Backend
WHAT WENT WRONG:
PNGs betrayed me: Stripped alpha channels and normalized Base64 so the backend stops puking 400s
Rate limits: auto-rotate API keys
Huge images: 413 errors so compressed and limited requests
PNG transparency to JPEG meltdown so forced RGBA to RGB
New slogan of CookieAI: Rage, Pain, Headache
Now I know the Gemini documentation by heart
For now, all good… ONLY FOR NOW
The top website for the top
See you soon :,)
http://159.65.96.245

DEVLOG #02 – Bug Fixes
This hour was very productive

Models not loading: The models were acting stubborn and refused to start. Fixed now
Responsive web: The website on mobile looked like a badly exported PowerPoint. Fixed! now is a godly exported powerpoint
AI glitching with bold text: Every time it tried to write in bold, it freaked out and spammed “**”
Design: Added some cleaner design practices

See you soon :3
http://159.65.96.245/ here the link

DEVLOG #01 – CookieAI
I built a simple AI web app (not official) for the dummies who have tons of questions about Flavortown.

Features:
OLED UI: Sleek, black-as-the-void
Triple-Threat AI Support: Integrated Gemini, OpenAI, and Claude. For those who want GPT or Claude, actually the web uses Gemini
Ez-Configurator: Classic Ez-Configurator I always build for my projects (dummy-proof)
Conversational Memory: Basic, but it actually works—burns tokens quickly #TODO

Technical Stuff:
Fallback Key System: Backend automatically rotates through multiple API keys if one hits a limit (like quota or something)

Notes:
The Gemini API had a weird system_instruction param error, so I moved the prompt to the first message in the history. Works better this way anyway

Web Version:
Right now it’s in a very basic version, but you can download the whole web, add your own API key, pick any model and prompt, and it will work for whatever purpose you want ask questions about your own website or anything else

Next Steps:
Responsive design for phones and other weird devices
Image support for multimodal AIs (Gemini will even support videos!)
Better animations

See you soon! ;3 ty
http://159.65.96.245/