llm-knowledge-agent

An agent to research a given topic by doing analytical and syntopic reading, summarizing and extracting main points, connecting them together, and forming new knowledge outputs such as zettels or evergreen notes.

Installation

Clone this repo and use poetry install.

Use

Before first use

1. Copy the example.env file to .env and fill out your keys.
2. Run load_knowledgebase command to create and store your evergreen note embeddings.

Creating a note

1. Create a textfile in this format for your article.
2. Run make_article --f FILE_LOCATION and your source and evergreen notes will be added to the directories you specified in .env

Warnings

• This is mostly a prototype to learn about building LLMs. With paid API access, this could get very costly very quickly and has not yet been optimized for cost-effectiveness.
• The current version is built for Obsidian-based knowledgebases, but there is an open issue for supporting additional tools for thought.
• This is also built around my own source and evergreen note templates. Feel free to rework the templates in llm_knowledge_agent/obsidian_templates.py to fit your templates. In the future this will be easier to configure.

Article file format

TITLE
AUTHORS (separated by commas)
TAGS (#format separated by commas)
URL
TEXT