Feature: Text-to-Cypher via LLM + RAG over schema

[peter7775]

Problem Statement

Users must know Cypher to query the graph via the GraphQL/REST API. For large schemas, the full schema also can't be sent as LLM context due to context window limits.

Proposed Solution

• Use semantic schema search to select a relevant schema subgraph as RAG context.
• Send the selected context plus the user's natural-language question to an LLM to generate a Cypher query.
• Validate/sandbox the generated Cypher before execution (e.g. only allow read-only MATCH queries, reject CREATE/DELETE/DETACH DELETE).
• Expose this as a new GraphQL mutation/query or REST endpoint.

Alternative Solutions

Send the LLM the entire schema without RAG selection — works only for small databases; larger ones exceed the context window and increase API costs.

Use Case

A non-technical user asks a question in natural language and gets a graph result without knowing Cypher.

Priority Level

Nice to have

Implementation Ideas

• Security review required before merging (see .gosec.json in the repo) given the risk of generating unsafe queries from untrusted input.
• Consider rate limiting / cost tracking for LLM API calls.

Additional Context

Depends on: "Semantic schema search", "EmbeddingProvider port". The most complex and highest-risk item in this set — recommend tackling it last.