OpenAI File Chat

This code implements a wrapper around OpenAI file search. It is convenient if you find the OpenAI interface clunky, or need to export its functionality without providing deep access to your OpenAI account.

Setup

Bash environment

Create the following environment variables:

$> ROOT=`git rev-parse --show-toplevel`
$> export PYTHONPATH=$ROOT
$> export PYTHONLOGLEVEL=info

Python environment

$> python -m venv venv
$> source venv/bin/activate
$> pip install -r requirements.txt

Configuration

Generate a configuration JSON. The structure is as follows:

{
  "gradio": {         dict # key-value pairs passed to Gradio.Blocks.launch
  },
  "openai": {
    "api_key":        str  # your OpenAI API key
    "model":          str  # OpenAI model to use
    "assistant_name": str  # Naming convention to give assistants
  },
  "chat": {
    "prefix":         str  # Vector store name prefix
    "system_prompt":  str  # Path to system prompt (static/system-prompt.txt)
    "retries":        int  # Number of prompt retry attempts
  }
}

Save this somewhere convenient, outside of this repository's directory structure.

Run the app

FILE_CHAT_CONFIG=/path/to/configuration.json gradio $ROOT/app.py

Navigate to the URL specified by Gradio, then open the accordion at the top of the page for usage instructions.