This is a fork of http://github.com/openai/chatgpt-retrieval-plugin that adds some improvements and resolves some issues. Follow the instructions in the README.md to install and run the plugin.
The upstream ChatGPT Retrieval Plugin binds to port 8000. The forked version allows you to use a custom port by setting the environment variable PORT. If unset, the port will be 8000. For example, to use port 8001:
export PORT=8001The upstream version uses a fixed chunk size of 200. The forked version's chunk size can be configured by setting the environment variable CHUNK_SIZE. If unset, the chunk size will be 200. For example, to set a chunk size of 1000:
export CHUNK_SIZE=1000In the upstream version, when using Weaviate and issuing a delete all request, the plugin will delete not only the documents but also the corresponding schema. In the forked version we resolve this by recreating the schema automatically after a delete all operation.
In order to run this in a quick prepackaged environment, together with GraphDB, you can use the bundled
docker-compose.yml. This will start GraphDB, Weaviate and the retrieval plugin on ports 7200, 8080 and 8000.
Remember to set your GPT key and the BEARER_TOKEN in the docker-compose.yml, as per our documentation.
Note that logs are set to INFO to prevent log spam. For debugging, please refer to the LOGURU_LEVEL variable.