Missing type mapping for vector(n) columns in PostgreSQL (pgvector)

[EmilioSanchisCHJ]

What problem are you trying to solve?

When running the EF Core Power Tools or the standard dotnet ef dbcontext scaffold command against a PostgreSQL database using the pgvector extension, the scaffolding process fails to map the vector data type.

As a result, columns intended for embeddings are skipped, and their associated indexes (such as HNSW) are not generated in the DbContext or entity models.

Steps to Reproduce
1.- Have a PostgreSQL database with the pgvector extension enabled.

2.- Create a table with a vector column, for example:

SQL
CREATE TABLE documento_resumen (
    id SERIAL PRIMARY KEY,
    embedding vector(768)
);
CREATE INDEX idx_doc_resumen_embedding_hnsw ON documento_resumen USING hnsw (embedding vector_cosine_ops);

3.- Run the scaffolding command
4.- Expected Behavior

The scaffolding engine should recognize the vector(n) type (likely mapping it to a Vector or float[] type if the Pgvector.EntityFrameworkCore.PostgreSQL library is present) and include the column and its indexes in the generated code.

5.- Actual Behavior
The following warnings are displayed, and the columns/indexes are omitted:

⚠️ Could not find type mapping for column 'public.documento_resumen.embedding' with data type 'vector(768)'. Skipping column.
⚠️ Unable to scaffold the index 'idx_doc_resumen_embedding_hnsw'. The following columns could not be scaffolded: embedding.
⚠️ Could not find type mapping for column 'public.expediente_resumen.embedding' with data type 'vector(768)'. Skipping column.
⚠️ Unable to scaffold the index 'idx_expediente_resumen_embedding_hnsw'. The following columns could not be scaffolded: embedding.

Describe the solution you'd like

• Proposed Solution
  Implement support for the vector data type within the scaffolding type mapper. Since pgvector is becoming a standard for AI-driven applications using PostgreSQL, native support for scaffolding these fields would significantly improve the developer experience for RAG (Retrieval-Augmented Generation) workflows.

[ErikEJ]

@roji @EmilioSanchisCHJ

What is the story here? Does Npgsql EF Core not support vector natively? I am reluctant to use third party mapper to enable this

[roji]

@ErikEJ that's right, as this is a PostgreSQL extension, this is supported via an EF type mapping plugin that's maintained by the same guy that does the PG extension. It's pretty standard though, I'd consider using it.

[ErikEJ]

Thanks @roji - so similar to spatial plugin?

[EmilioSanchisCHJ]

Yes, it's similar to the spatial plugin.

[ErikEJ]

Ok I will investigate making it an option similar to spatial and nodatime.

[EmilioSanchisCHJ]

In my case, the class that has worked perfectly for me is:

// <auto-generated> This file has been auto generated by EF Core Power Tools. </auto-generated>
#nullable disable
usingSystem;
using System.Collections.Generic;
using System.ComponentModel.DataAnnotations;
using System.ComponentModel.DataAnnotations.Schema;
using Microsoft.EntityFrameworkCore;
using NpgsqlTypes;
using Pgvector;

namespace Scriptum_v4.ModelsPostgress;


public partial class documento_resumen
{ 
[Key] 
public int id { get; set; } 

[Column(TypeName = "vector(768)")] 
public Vector embedding { get; set; } 
}

The vector(768) is the number of dimensions of the Vector in the database

[EmilioSanchisCHJ]

And in OnModelCreating:

entity.Property(e => e.embedding)
       .HasColumnType("vector(768)");