Can I embed my GraphQL schema into a vector database like Pinecone and get LLM to generate the query/mutations?

Question

I am experimenting with a new UI where typical profile settings are updated via chat instead of UI. For example, instead of showing a frontend component to let users cancel their billing, they can just chat to the bot.

I am wondering if its possible to get my LLM (lets say gpt3) to generate the graphql query necessary to run these operation. I am thinking I could ingest my graphql schema into a vector database like Pinecone, then feed in that context into LLM so that it can generate the appropriate GQL query/mutation.

Is this feasible, or a bad idea?

I have only theorized this so far

Did you find any solution? I am also trying to solve a similar problem and this feels like a good way to fit larger schemas inside models' context limit. — May 26 at 16:07

user6714507user6714507 631 silver badge9 bronze badges · Accepted Answer · 2023-06-19 14:25:50Z

0

I managed to do this on simple queries by using Langchain’s graphql agent

answered Jun 19 at 14:25

user6714507

631 silver badge9 bronze badges

0

score 0 · Accepted Answer · 2023-09-12 15:05:59Z

this is not an answer.

I’m working on a similar issue, I was wondering if we can use gql introspection query to get just "the field needed" to formulate the query. I’ve tried using gql agent, but I find on more complex schema, it does hallucinate non-existant field even when schema is provided as context.

Can I embed my GraphQL schema into a vector database like Pinecone and get LLM to generate the query/mutations?