Vector field type and similarity search

New Feature / Enhancement Checklist

• Report security issues confidentially.
• Any contribution is under this license.
• Before posting search existing issues.

Current Limitation

There is no way to store embeddings or run a similarity search through Parse Server. If you are building anything AI related you have to bypass Parse and query the database directly, which skips ACLs, CLPs, triggers, etc.

Feature / Enhancement Description

A Vector schema field type and a query constraint like $nearVector that returns nearest neighbours sorted by similarity. Implemented per storage adapter; both backends have native support (MongoDB Atlas Vector Search, pgvector on Postgres), so this fits the same pattern we already use for full text search ($text).

Example Use Case

1. Add a Vector field to a class, e.g. embedding with a dimension
2. Save objects with embeddings in that field
3. Query with $nearVector and get results ranked by similarity, with normal query constraints and ACLs still applying

Alternatives / Workarounds

• Query the database directly (bypasses the Parse security layer)
• Run a separate vector database next to Parse and keep it in sync manually

3rd Party References

• MongoDB Atlas Vector Search
• pgvector (used by Supabase)
• Firestore vector search

No specific files or tests are named. Start by examining the existing full-text $text search pattern and how storage adapters expose it, then review MongoDB Atlas Vector Search and pgvector constraints; done means vector fields and similarity queries work across both backends while preserving normal constraints and Parse security.