Skip to content

bramses/chatgpt-md

Repository files navigation

ChatGPT-MD

--> APRIL 2024 LOOKING FOR NEW MAINTANERS (SEE HERE). DM ME IF THAT'S YOU! <--

A (nearly) seamless integration of ChatGPT into Obsidian.

Demo

Screen.Recording.2023-03-05.at.9.16.31.PM.mov

(youtube mirror - for mobile users ⬇️)

video thumbnail

Features

  • Chat from any MD note
  • Create Chat Templates for sharing and running similar scenarios. Check out the companion repo chatgpt-md-templates for some templates!
  • As minimal boilerplate as possible, only two required in fact! <hr class="__chatgpt_plugin"> and role::system|assistant|user
  • Use frontmatter to change variables for the ChatGPT API
  • Stream characters to Obsidian, creating a realtime feel
  • Uses regular Markdown. Meaning everything from lists to code blocks from ChatGPT will render!
  • Create chats from highlighted text.
  • Infer title from messages. Can be set to run automatically after >4 messages.
  • Stream at cursor position or at end of file. Can be set in settings.
  • Choose heading level for role h1-h6. Can be set in settings.
  • Custom endpoints can be specified using the url parameter in your front matter. See FAQ for an example.
  • Stop a running stream with a command. See commands section below.
  • (NEW!) Choose between nine languages for "Infer Title". Can be set in settings.
  • (NEW!) ChatGPT comment blocks. Allows you to leave scratchpad notes, backlinks...or anything else really!! See command below for details.

Commands

Chat

The main command! Parses the file and calls ChatGPT. Recommended to add to a hotkey for easy usage.

Create New Chat with Highlighted Text

Take currently highlighted text and default frontmatter and create a new chat file in Chat Folder

Create New Chat From Template

Create a new chat file from a template specified in Chat Template Folder. Remember to check out chatgpt-md-templates for some templates!

Infer Title

Infer the title of the chat from the messages. Requires at least 2 messages. Can be set in settings to run automatically after >4 messages.

Add comment block

Add a comment block to the editor that will not be processed by ChatGPT. Allows you to leave scratchpad notes, backlinks...or anything else really!

Comments begin with =begin-chatgpt-md-comment and end with =end-chatgpt-md-comment

Screenshot 2023-04-03 16-47-05 Screenshot 2023-04-03 16-59-26

Clear Chat

Removes all messages but leaves frontmatter

Stop Streaming (Does not work on mobile)

Stops the stream. Useful if you want to stop the stream if you don't like where ChatGPT is heading/going too long.

Add Divider

Add a ChatGPT MD Horizontal Rule and role::user.

!! Note: both role::system|assistant|user AND <hr class="__chatgpt_plugin"> are REQUIRED for the plugin to work!!

Installation

Community Plugins

Go to Community Plugins and search ChatGPT MD

Local

  1. Clone this repo into your plugins directory in your vault
  2. Run npm i and npm run build

Both

  1. Insert your OpenAI API Key into the settings
  2. Set Chat Folder and Chat Template Folder
  3. Add a hotkey for Chat (Im using alt-[)

FAQ

Q: The chat seems to be getting cut off halfway through

To address this, first try to increase your max_tokens (default is set to 300). You may also want to update it more permanently in the default frontmatter settings. See pics below:

Screenshot 2023-03-12 16-14-35 Screenshot 2023-03-12 16-15-01

Q: Code Blocks cut off halfway through and leave ```

The Obsidian editor renders backticks in automatically (see issue) and fires extra logic that causes the stream to add extra backticks. To address this, you can:

  1. at the end of the code block add ``` (three backticks) to close the code block BEFORE the <hr> and delete the three extra Obsidian added automatically.
  2. in role::user write "keep going"

See pics below:

Screenshot 2023-03-15 18-47-40 Screenshot 2023-03-15 18-48-30

Q: How do I use GPT-4?

If you are off the waitlist, simply replace model: gpt-3.5-turbo with model: gpt-4 in the frontmatter. (note: gpt-4 is slower than turbo!)

Q: How do I use a custom endpoint?

---
system_commands: ['I create small self contained app ideas that could fit in a CodePen or a Replit']
url: https://localhost
---

The custom API must conform to the OpenAI API spec. eg you could use Azure's OpenAI hosted endpoints here. Refer to your provider for API key handling.

About the Developer

This repository was written by Bram Adams, a writer and programmer based out of NYC.

Bram publishes a weekly newsletter, is a community developer ambassador for OpenAI, and does freeleance contracts (for hire!) related to AI/web dev/AR+VR.

As of Fall 2023, Bram is actively working on Commonplace Bot, a novel and modern look into how LLMs and creative coding can help us capture, engage, and creatively remix the best pieces of information we come across. You can test it out live in the Bram Adams' Discord server.

Bram is the creator of Stenography, an API and VSC Extension that automatically documents code on save, which went #1 on Product Hunt. He also is the author of Bramses' Highly Opinionated Vault, an extremely detailed philosophy + vault template used by thousands of Obsidian users (new and old!), and ChatGPT MD, a (nearly) seemless integration of Chat GPT into Obsidian which has been downloaded by over twenty thousand Obsidian users. He also taught the GPT-3 in Production Course for O'Reilly Media, teaching students how to leverage LLMs in the real world of production.

Previously Developer Advocate @ Algolia, Software Engineer @ HBO, Computer Science B.S. @ University of Rochester

You can learn more about him and his work on his website.

The best way to support his work is to sign up for his newsletter here.