Glasp is highly relevant to the AI agent ecosystem because it focuses on the data capture and retrieval problem that plagues many personal assistants. By providing a structured way to highlight and tag web content, they are effectively creating a high-quality, human-curated training set for personal RAG (Retrieval-Augmented Generation) systems. Their "AI Clone" feature is a direct application of this, allowing an LLM to act as an agent over a specific user's reading history.

In the broader agent stack, Glasp acts as a bridge between the browser (where knowledge is consumed) and the model (where knowledge is processed). Their work on YouTube summarization and PDF highlighting provides the necessary parsing to turn unstructured web media into structured, agent-readable data. For developers building personal agents, Glasp represents a potential source of clean, annotated user preference data that goes beyond what can be inferred from simple browsing history.

Most bookmarking apps are digital graveyards. We save links with the intention of reading them later, only for those links to vanish into an unorganized list that is never revisited. Glasp takes a different approach by focusing on the act of highlighting and the social visibility of those notes. Founded in 2021 by Kazuki Nakayashiki and Kyosuke, the San Francisco-based company builds a browser extension that overlays a highlighter tool directly onto the web. Unlike traditional read-it-later apps that strip a page into a clean text view, Glasp works on the live site, allowing users to color-code snippets of articles and YouTube transcripts in real-time.

The core product mechanics are straightforward. Users install the extension and begin highlighting text. These highlights are then aggregated on a personal profile page, which functions as a public-facing knowledge base. This social aspect is the primary differentiator. While tools like Readwise focus on personal retention and syncing to apps like Notion or Obsidian, Glasp is built for discovery. Users can follow others to see what they are reading and, more importantly, which specific parts of a long-form article they found valuable. This creates a signal-to-noise filter powered by human curation rather than an algorithm.

In the last year, Glasp has aggressively integrated generative AI into this workflow. They are perhaps best known for their "YouTube Summary with ChatGPT & Claude" extension, which allows users to get an instant summary of a video transcript. This feature solves the friction of consuming long-form video content and provides a structured way to extract insights. Beyond simple summaries, the company is experimenting with "AI clones." By using a user's library of highlights and notes as a training set, the platform can generate a digital persona that answers questions based on that specific individual's knowledge base. This moves Glasp from a simple curation tool to a personal intelligence platform.

From a competitive standpoint, Glasp occupies the space between social networks like X and utility tools like Pocket. It is less about broad broadcasting and more about domain expertise. The company remains small, with a team of fewer than ten people, and has raised a pre-seed round of approximately $400,000. Their primary challenge is the tension between public sharing and the private nature of deep research. While many researchers prefer private silos, Glasp bets that the future of knowledge work is collaborative and that the data generated through highlighting is the perfect fuel for personalized AI models. By capturing the "why" behind a saved link—the specific sentence or idea that resonated—they are building a dataset that is significantly more valuable than a simple list of URLs. The platform is currently free to use, focusing on user growth and the accumulation of this high-intent knowledge graph.