Pascal, my article-saving agent
How I actually read the articles I save. I send a link to Pascal, my Hermes agent, on Telegram. It files the article into an Obsidian vault, tags it and replies with a summary in under a minute.
- Why
- About 98% of the long Substack, LinkedIn and X articles I saved went to a 'saved' graveyard and were never opened again. Now the summary tells me whether to keep reading, and the vault has become a searchable map of what I've been learning.
- How long
- Running daily for 100 days by July 2026
How I built it
- I installed Hermes Agent on my Mac and connected it to Telegram, so I could message it like a person.
- I pointed it at a local model in Ollama.
- I wrote a skill that scrapes a shared link, files the article into my Obsidian vault, labels and tags it, then replies with a short summary.
- I made it a habit: whenever I want to save an article, I share it to Pascal instead.
- After 100 days I had the AI turn the vault into Pascal's Nebula, a star map of my saved articles grouped by theme.
- To copy it you need
- An always-on agent you can message (Hermes Agent, or the AI of your choice with a Telegram bot), Obsidian, and a model to summarize. Gotcha: some sites block scraping, so ask the agent to say clearly when it could not read an article instead of summarizing the headline.
Try it yourself
Copy this prompt and paste it into ChatGPT or Claude to try it yourself.
I save too many articles and never read them. I want to copy Pascal, the article-saving agent Korab Eland describes on whatarewebuilding.ai (korabeland.com/notes/pascals-nebula). Korab shares a link on Telegram, and the agent scrapes the article, files it into an Obsidian vault with tags, and replies with a summary in under a minute. If the summary is interesting, Korab keeps reading. Help me build a starter version. Walk me through the simplest setup for my phone and computer, then write the instructions the agent follows: what to save, how to tag it, and how long the summary should be. Have it say plainly when it could not read a page. Start with the messaging part. Ask me for anything you need.
Tools I used for this project
