Docs Chat

Docs Chat is a UI interface will be a UI interface for langchain's document loader embedding wrapper, used to fit Documentation, Books, or any other kind of "Document" into retrieval context for an LLM. I'm using it here with GPT-3.5 to "talk" with documentation I don't really understand. Usage currently requires a developer's understanding of Node projects and environment variables.

Usage

With A Non-existent incomplete UI

This SvelteKit project uses node version 18 as a dependent version for langchainJS. Usage follows standard convention for other SvelteKit projects.

Install dependencies needed to run this project with

 npm install

and run this project with

 npm run dev

In its current state

This project requires an OpenAI API key to run. You can find API Key information here. In this particular hardcoded environment, I'm referencing the API key from my own .bashrc file inside queryFile.js, and as such this assumes your terminal has an exported OPENAI_API_KEY variable.

Currently, this project's whole functionality is limited to the src/utils/ingestFile.js file, and src/utils/queryFile.js files. the ingestFile.js file is used to convert a text file into an embedding that can be used by langchainJS.

You can load a file to ask a model questions about with

npm run ingest <file-name> <output-directory-name>

assuming you've placed the file in the src/data/text directory, and <file-name> is the name of the file you want to embed, and <output-directory-name> is the name of the directory you want to save the embedding within the data/docs directory. This only needs to be done once per file, and the embedding will be saved in the data/docs directory for future use.

Important Disclaimer

Ingesting a file uses OpenAI's embedding API, and as such, will cost you literal actual money. Using other models besides GPT-3.5.turbo will also rack up non-negligible costs.

The queryFile.js file is a hard coded example of how to use langchainJS to ask questions about your embedded text file. Modify the prompts as you see fit, change the relevant file names/directory names, and see the output with node src/utils/queryFile.js in your terminal.