Goal
Give users a built-in personal Knowledgebase for organizing user-specific knowledge in a structured format and querying it naturally through prompts.
One of the “whys” is also to let the chats in the ChatGPT history stop bliping out of existence and provide some well defined outcome formats.
Monetization
It is a great feature for paid users, or even as a feature package that one could pay for.
Core ideas
-
Organize user-specific knowledge as structured Content Types instead of keeping everything as unstructured chat text.
-
Content-types are available in UI as independent menu entries near the top left New Chat option.
-
User can interact with Content-type as through UI as well as through chats.
-
Let users create their own Content Types and define their, for example (those maybe bootstraped for new users if they wish):
-
TODO
-
Ideas
-
Wishlist
-
Shopping List
-
Books
-
Phrases
-
Links
-
Papers
-
-
Each Content Type can define:
-
Schema — fields and metadata (YAML-like jsonschema?)
-
Skill — how ChatGPT should work with this type
-
-
Allow users to query and modify the Knowledgebase through normal prompts.
-
Conten-type skills are supported by internal OpenAI query API for accessing the user’s structured Knowledgebase. (e.g. how to query it and how it will happen in backend)
Users could then ask:
-
“Show what I planned to do”
-
“Add this coffee machine to my wishlist.”
-
ChatGPT work: “Please check regularly my wishlist and inform me about discounted offers”
-
The core model would be:
Core entries would be:
- Knowledgebase
- Content Types which defined
- Structured Items of Conten-Types
- Skill that is exposed from Content-Type
- Internal KB query API
There also could be a special field “body” which will have a freeform text editable as document through UI.
This would make ChatGPT useful for structured, queryable personal knowledge system.
Idea is great and works locally pretty-well 