Why should you care?

One database for accounts plus vectors simplifies backups and transactional consistency for small tenant KBs.

See it live — copy this example

Paste into an ASP.NET Core 8+ / AgentNest project, then run with dotnet run (set your API keys in user-secrets).

// AgentNest.Infrastructure/PgVectorStore.cs
public sealed class PgVectorStore(NpgsqlDataSource ds)
{
    public async Task UpsertAsync(string tenantId, string id, float[] vector, string text, CancellationToken ct)
    {
        await using var cmd = ds.CreateCommand("""
            INSERT INTO kb_chunks (id, tenant_id, embedding, content)
            VALUES ($1, $2, $3, $4)
            ON CONFLICT (id) DO UPDATE SET embedding = EXCLUDED.embedding, content = EXCLUDED.content
            """);
        cmd.Parameters.AddWithValue(id);
        cmd.Parameters.AddWithValue(tenantId);
        cmd.Parameters.AddWithValue(new Pgvector.Vector(vector));
        cmd.Parameters.AddWithValue(text);
        await cmd.ExecuteNonQueryAsync(ct);
    }
}

What happened?

• PgVectorStore upserts into kb_chunks with pgvector type.
• CRM relational data and embeddings share Postgres backups.

Practice next

1. CREATE EXTENSION vector; on AgentNest Postgres.
2. dotnet add package Pgvector and Npgsql.
3. CREATE INDEX ON kb_chunks USING hnsw (embedding vector_cosine_ops);
4. Add generated tsvector column for hybrid keyword + vector search.
5. Partition kb_chunks by tenant_id for large SaaS.

Remember

pgvector colocates vectors with CRM/ERP Postgres data. Use HNSW index for production similarity search. Always filter tenant_id in SQL, not only in app code.

Unified CRM database

Team refuses separate vector SaaS; CRM rows and playbooks live in Postgres.

Outcome: pgvector kb_chunks table powers copilot RAG with single backup policy.