GraphQL Embeddings: AI-Powered Dynamic Operations From Schema To IDE
summary
An open-source toolkit called GraphQL Workbench parses a GraphQL schema into chunked documents, embeds them with embedding Gemma 300, stores vectors in PG Lite/Postgres/Pinecone, and uses a multi-agent LLM pipeline (default Qwen 2.5 via Ollama) to generate validated GraphQL operations. Small schema: 296 documents embedded in ~5s; mid schema: 1,259 documents in 7.5s; very large graphs take ~7 minutes to embed. Operation generation runs locally (Jetson Orin 8GB: 60–90s; another demo 18.4s), includes an MCP validation loop and incremental diffs, and ships as a VS Code extension under MIT.
AI-written summary · about
chapters (8)
- 0:00Introduction and motivation
- 3:31GraphQL embeddings concept
- 7:24Live demo setup
- 10:11Generating operations
- 13:26Operation generation process
- 17:58Schema design features
- 21:48Architecture and extensibility
- 24:49Q&A on models
related talks
29mClosing the Loop: How GraphQL Gives Coding Agents Eyes on What Actually Matters
35mDynamic, Safe Ops Using AI to Generate Trusted Operations from Text Prompts
35mGraphQL Docs: Beyond the Schema
30mGraphQL Field Discovery & Query Generation with Generative AI
30mSponsored Session: GraphQLs Future is Rooted in Governance
24mGraphQL in Production: Empowering Efficient Data Retrieval and Unleashing Developer Productivity
26mJust Ask: Building a Natural Language GraphQL Agent from the Graph You Already…
26mSponsored Session: From Prototype to Production: Building Custom GraphQL APIs