Following up on the rewrite of the Browser Recall server, which I wrote about here, I've now implemented a weekly report using that data.
Browser Recall Weekly will take your last week of browsing data and generate a full report with insights on the major themes of your browsing history for the last week. This works by first going through each records markdown and generating a short summary, list of key topics, and assigning a category. That data is then used to generate the larger report along with an interactive chart and a pretty word cloud.
The idea is to allow me to reflect on my activity each week and potentially gain some insight into the higher level themes I may not be aware of. Additionally once I have a bit more historical data I can do some higher level analysis on where my web journey takes me over time.
This is all done using local models, specifically Gemma 3 12B. It took about an hour to get through the last week of data, but the time will entirely depend on how much content was in the pages you visited, as there is a chunking process to accomodate a 4096 token context window.
I'm looking forward to exploring more low urgency/high compute tasks like this now that I have a machine I can run an LLM on 24/7.
Here's the report for my first week:
