bookmark_assistant

provenance:github:KianoushAmirpour/bookmark_assistant

WHAT THIS AGENT DOES

This agent helps organize your Chrome bookmarks. It automatically sorts them into categories like "News," "Shopping," or "Travel," and it can also reorder them to show the most recently added ones first. This solves the problem of having a cluttered and hard-to-navigate collection of bookmarks. Anyone who uses Chrome and has a lot of saved links would find this helpful, especially those who struggle to keep their bookmarks tidy. What makes it special is that it learns and improves over time based on feedback, ensuring your bookmark organization gets better and better. The agent presents its work in a clear, easy-to-read format.

View Source ↗First seen 5mo agoNot yet hireable

README

## AI Agents for Managing Chrome Bookmarks
This repository explores using AI agents to enhance the management of Chrome bookmarks.  
It contains two main agent workflows—**one for classification and one for recency ranking**—along with supporting utilities, evaluation data, and a reflection-based improvement loop.  

## Overview
This project implements AI agents designed to:  
* Classify bookmark URLs into meaningful groups with descriptive category names.  
* Rank bookmarks based on when they were added, enabling recency-based organization.  

## Utilities & Tools
* Converting WebKit timestamps to standard datetime formats.
* Sorting lists by key.
* Formatting timestamps into human-readable strings.
* Reading and writing JSON and Markdown files.

## Reflection Pattern
Both agents employ the reflection pattern, where the model receives feedback— **LLM-as-judge or from tools**.  
Classification Agent: Reflection via rubric-based grading. [prompt for reflection](https://github.com/KianoushAmirpour/bookmark_assistant/blob/main/prompts/classifier_reflection_prompt.py)   
Recency Agent: Reflection powered by external, tool-assisted feedback. 

## Evaluation
A dedicated evaluation dataset has been created to test the recency ranking agent.

## Tech stack
* pydantic-ai and logfire

## Outputs 
All outputs are in markdown format. They will be saved in outputs folder.

## to do
* support for mac and linux
* prompt versioning
* add logging
* eval dataset for classifier agent

PUBLIC HISTORY

First discoveredMar 22, 2026

IDENTITY

inferred

Identity inferred from code signals. No PROVENANCE.yml found.

Is this yours? Claim it →

METADATA

platformgithub

first seenNov 7, 2025

last updatedNov 12, 2025

last crawled18 days ago

version—

README BADGE

Add to your README:

![Provenance](https://getprovenance.dev/api/badge?id=provenance:github:KianoushAmirpour/bookmark_assistant)