I recently discovered the Fabric project on Github. The gist of that project is to collect an open-source set of LLM prompts, share them among the public and integrate them with other tools.

This week, I gave it a go within my Obsidian.md vault, with the help of a plugin called MeshAI. Suffice it to say, my workflow has been significantly impacted by GenAI once again.

It works in four steps:

  • Configure an LLM provider
    • I like LMStudio, I run it locally on my computer and use whichever latest model I’m enjoying
    • As of this week, my local LLM workhorse is Microsoft Phi 4
  • Set your input context
    • Input can come from an existing Obsidian Note
    • Or, the really cool option, from a Tavily Search
  • Select which “patterns” to apply to the given input
    • Extra cool note, these patterns can be stitched together to work like an agentic system!
  • Write the output document

Putting that all together, I have built myself a mixture of LLM agents, that have access to web research and can provide me with exactly the results I want based on my “patterns”.

Check out this example, this is insane:

In this example, I’ve emulated a deep-research style feature, and built it into my Obsidian vault.

  • Local phi4 LLM ensures data doesn’t leave my machine
  • Tavily search provider gives my local LLM access to gather data from the internet
    • I just type the subject of interest that I want to research
  • 3 custom “patterns” work together like an agentic system to provide me desired output

That all took a few minutes to run, cost me nothing, and produced the rather outstanding research note below.

Transclude of Agentic-Phi4-Summary--Logarithmic-Regressions