docsgpt's Introduction

DocsGPT

Quick insights from docs & references. Given any URL and a question DocsGPT will respond using the document as context.

Setup

cd DocsGPT
export OPENAI_API_KEY="sk-..."
chmod +x ./docs-gpt
./docs-gpt --start-server

Examples

Google SRE Chapter

./docs-gpt https://sre.google/sre-book/production-environment/ --question "What is Borg?"

Output

Response: Borg is a cluster management system developed by Google for running large-scale distributed workloads.

In this example the GPT Agent fetches the results of the url https://sre.google/sre-book/production-environment/' and is asked the question: What is Borg?

Attention Is All You Need (PDF)

./docs-gpt https://proceedings.neurips.cc/paper_files/paper/2017/file/3f5ee243547dee91fbd053c1c4a845aa-Paper.pdf --question "What is multi-head attention?"

Output

Response: Multi-head attention is a type of attention mechanism used in neural networks that allows the model to attend to different parts of the input sequence simultaneously. It is a form of self-attention that allows the model to focus on different parts of the input sequence at the same time.

In this example the GPT Agent is able to fetch and parse the PDF and answer the question pertaining to Multi-Head Attention in the context of this famous Google Research Paper.

REPL

Start REPL

./docs-gpt --start-repl

REPL

Enter :q to quit repl

Enter URL: example.com/confusing-documentation

What question do you have?: How does this confusing thing work?

TODO

Add CLI
Add REPL
Add FastAPI server
Create a UI to interact with the agent
Compare portions of text using document comparison and grab the most relevant bodies of text to the given question.
Store document embeddings and create 'sessions'

Recommend Projects