ragserver

RAG stands for Retrieval Augmented Generation

Demos of implementing a “RAG Server” in Go, using Google AI for embeddings and language models and Weaviate as a vector database.

How it works

The server we're developing is a standard Go HTTP server, listening on a local port. See the next section for the request schema for this server. It supports adding new documents to its context, and getting queries that would use this context.

Weaviate has the be installed locally; the easiest way to do so is by using docker-compose as described in the Usage section.

Server request schema

/add/: POST {"documents": [{"text": "..."}, {"text": "..."}, ...]}
  response: OK (no body)

/query/: POST {"content": "..."}
  response: model response as a string

Server variants

• ragserver: uses the Google AI Go SDK directly for LLM calls and embeddings, and the Weaviate Go client library directly for interacting with Weaviate.
• ragserver-langchaingo: uses LangChain for Go to interact with Weaviate and Google's LLM and embedding models.
• ragserver-genkit: uses Genkit Go to interact with Weaviate and Google's LLM and embedding models.

Usage

• In terminal window 1, cd tests and run docker-compose up; This will start the weaviate service in the foreground.
• In terminal window 2, run GEMINI_API_KEY=... go run . in the tested ragserver directory.
• In terminal window 3, we can now run scripts to clear/populate the weaviate DB and interact with ragserver. The following instructions are for terminal window 3.

Run cd tests; then we can clear out the weaviate DB with ./weaviate-delete-objects.sh. To add documents to the DB through ragserver, run ./add-documents.sh. For a sample query, run ./query.sh Adjust the contents of these scripts as needed.

Environment variables

• SERVERPORT: the port this server is listening on (default 9020)
• WVPORT: the port Weaviate is listening on (default 9035)
• GEMINI_API_KEY: API key for the Gemini service at https://ai.google.dev