I’m a regular college student who practically lives on LLMs. (I routinely max out 2 Gemini Pro accounts and 2 Cursor Pro accounts every single day just to keep up.)
As my conversation volume grew, I got incredibly frustrated by the LLM "amnesia"—having to re-explain my background and context from scratch over and over again. I finally lost my patience and built a solution tailored exactly to my own needs.
Originally, this was strictly for personal use. I built a prototype in 2 days and used it, and my own experience became so seamless that an irrational sense of confidence and baseless arrogance (?) kicked in: the world is missing out if I don't publish this. So I decided to release it.
I’m an applied statistics student, not a professional developer, and this started as a hobby project. But I spent the last week burning through 4 Cursor Pro accounts at maximum capacity, obsessively polishing edge cases, concurrency issues, and data integrity details. Instead of just blindly dumping all chat history into the prompt, I built a selective retrieval pipeline using SQLite and Pinecone, structured into 3 layers: Tier 1 (Global Identity), Tier 2 (Workspace Knowledge), and Tier 3 (Vector Archive).
Personally, I believe this currently provides a more satisfying personal user experience than any other LLM interface out there right now.
I’m covering the server costs myself, so please check out the resources below and give it a spin.
I wrote this with a lot of confidence, but if you look under the hood, it might just look like a clueless college student's scrappy implementation. Still, spending a week densely living through brainstorming, fixing various edge cases and bugs, and obsessing over my own details was an incredibly rewarding experience. Just take a look. You won't regret it.
If you end up using it, you can send feedback through the settings menu, so please send a lot of it! (I'll try my best to answer after I survive my final exams... doing random things during exam week is the ultimate procrastination...)
P.S. My English isn't perfect, so I used AI to help translate this post and the docs. If any nuances got lost in translation or sound a bit off, please bear with me!
dominicyglee•1h ago
I’m a regular college student who practically lives on LLMs. (I routinely max out 2 Gemini Pro accounts and 2 Cursor Pro accounts every single day just to keep up.)
As my conversation volume grew, I got incredibly frustrated by the LLM "amnesia"—having to re-explain my background and context from scratch over and over again. I finally lost my patience and built a solution tailored exactly to my own needs.
Originally, this was strictly for personal use. I built a prototype in 2 days and used it, and my own experience became so seamless that an irrational sense of confidence and baseless arrogance (?) kicked in: the world is missing out if I don't publish this. So I decided to release it.
I’m an applied statistics student, not a professional developer, and this started as a hobby project. But I spent the last week burning through 4 Cursor Pro accounts at maximum capacity, obsessively polishing edge cases, concurrency issues, and data integrity details. Instead of just blindly dumping all chat history into the prompt, I built a selective retrieval pipeline using SQLite and Pinecone, structured into 3 layers: Tier 1 (Global Identity), Tier 2 (Workspace Knowledge), and Tier 3 (Vector Archive).
Personally, I believe this currently provides a more satisfying personal user experience than any other LLM interface out there right now.
I’m covering the server costs myself, so please check out the resources below and give it a spin.
I wrote this with a lot of confidence, but if you look under the hood, it might just look like a clueless college student's scrappy implementation. Still, spending a week densely living through brainstorming, fixing various edge cases and bugs, and obsessing over my own details was an incredibly rewarding experience. Just take a look. You won't regret it.
If you end up using it, you can send feedback through the settings menu, so please send a lot of it! (I'll try my best to answer after I survive my final exams... doing random things during exam week is the ultimate procrastination...)
P.S. My English isn't perfect, so I used AI to help translate this post and the docs. If any nuances got lost in translation or sound a bit off, please bear with me!
Thanks.
Service Link : https://dl-chat-49232436682.asia-northeast3.run.app/ Overview & Technical Whitepaper : https://gist.github.com/Dominic-YG-Lee/d49861c8bc9889bd8c946...