Inspiration
If agents could have access to the same human context window over my past 15 years of Internet usage, it could lead to interesting research and product implications.
What it does
Archives all web pages you visit, not just saving hyperlinks. This becomes your own personal training data.
How we built it
Cursor
Challenges we ran into
Done with archive chromium plugin. Began working on index and semantic search. There are privacy, performance, and storage size concerns.
Accomplishments that we're proud of
We think a few research scientists I spoke to confirmed that this would be a productivity boost. I think extended browser history is important for two reasons: 1) if I want to train my own personal LLM in near future, my own corpus of multi-modal data would be important; 2) extended browser memory as well as captured HCI keystrokes should have interesting implications to training Jarvis today and agentic behaviors in the future.
What we learned
This is more than a 24hr hackathon project.
What's next for Extended Browser Memory
Do people want this? I get annoyed from my own imperfect memory. My intent is not to win a hackathon, I am genuinely interested whether xAI team finds this approach useful / helpful product idea.
Log in or sign up for Devpost to join the conversation.