> [!WARNING] Setup Required > I disabled the plugin in the vault for now, if you want to use it just enable it in the settings > [!TIP] Video Example > ![Video](https://youtu.be/xHY3OHUNOrQ?si=_isujrZbJpYTJN2T) ## How it works - First, the notes in your vault get broken up into different blocks based on its headings. - Next, to determine the relevance of each block, each one is sent to OpenAI, which generates an embedding for that particular block, a set of values that describe its characteristics across different dimensions. In this case, it's described across different dimensions (for reference, [[#OpenAI]] small embeddings are 1536) - ![](https://i.imgur.com/iglEJOZ.jpeg) - Once the embeddings are calculated for each block, they are stored in a hidden file within your Obsidian vault. - So when you are asking the app to find similar connections to a block, it will find other blocks with similar values for each of the parameters, and this relevance is expressed through a percentage. - It's like finding the nearest dot on a graph, except the graph is 1536 dimensions big. - This is really powerful since it can find similarities in dimensions that we might not be consciously thinking of, allowing for even crazier connections ## Features ### Customization #### Included notes - Can exclude/include notes for different contexts - Maybe you are looking for everything so you keep it set to all - Maybe you are looking for certain memories so you are looking for #### Embedding type *Found in "Embedding Models" section of settings* Two main ways embeddings can be generated ##### Local - Less dimensions to measure your notes - Cheaper as you don't have to pay OpenAI per embedding you generate - Requires good device specs. If your specs aren't good enough for the embeddings you are trying to generate it will continuously fail to generate. ##### OpenAI - More dimensions (I haven't done an objective comparison but when I use [[#Relatedness View]] the results feel more specific) - Can be much more sizeable (local to real is like 10x) - Brian recommends to use the small local model - Costs money to generate - 165,000 word and 460 notes - $0.17 USD to create embeddings - Can see pricing [here](https://platform.openai.com/docs/guides/embeddings) You will need to set up your OpenAI API key - If you don't know how to get one, you can create an OpenAI account then go to https://platform.openai.com/account/api-keys - Create a new secret key, then copy and paste it into the vault settings ##### Opinions It seems like the price for generating embeddings is low so I personally generate embeddings. If you use multiple devices with different specs, using online embeddings might be better ### Relatedness View - Can see what sections of your notes are most relevant ### Chat View - Can use your vault as the database for notes - Can save chats just like ChatGPT ## Wishes - Quote embeds would be part of the note, since it's a core part of my input process - I don't think it's worth yet as I can just copy paste the content of the note, but it is unfortunate lol