LLMs like markdown, humans like markdown.

The idea is to build a “computer” or a framework and set of apps that are all based in markdown so that a LLM can operate it as well as a human can. The software will also include a CLI so the LLM and Human can run commands

The system prompts set up for the LLM will include lots of system data, including the majority of the main directory as to remain efficient

The LLM can be whatever you choose, using OpenAI compatible APIs

The LLM will have a dedicated folder for it to write notes and instructions for itself in

Each file will include some metadata meant for the LLM, not the human, including generated summaries and notes, as to make finding files and interpreting them easier and more token efficient for the LLM

Eventually, this software will include email, some messaging apps, calendar, web browser, and of course word processor with the majority of the interface and content being in markdown

IDK how images and videos are going to work yet

github repo here: https://github.com/findirfin/markdown-computer I may post updates here: https://findirfin.dev/md/posts/markdown-computer/

Log#

just got a webview to show up! I’m very happy about this. this will make it pretty easy to make the rest of the GUI

I got a basic layout implemented need to remember that the system prompt needs to include the page currently open

finally got it to populate the file list!!!

IF I CONVERT EVERYTHING TO TEXT (MARKDOWN) LLMs ARE AGI!!!!!

folders and nested folders (almost) working!

this update is basically just git except without the code

Folders are now collapsible and expandable and it finally works

actually got git set up lol

made the file explorer show what files are open. I also did a lot of work on the structure of the projects, so im basically at the same point as i was earlier, but expanding the project will work much better from now on. the file structure updates every 10 seconds.

to do (for the file explorer): add functions to make files and folders with python connect the html and js to this and add a gui for it. make a open file function in python and connect it to html (so the ai can use it to open files later) improve the gui for the x button for components