🐥 Vector embeddings with Ash, OpenAI, and PostgreSQL

Ash Framework makes it straightforward to integrate machine learning into your Elixir applications. With AshAi and PostgreSQL’s vector extension, you can embed text fields and make them ready for semantic search, recommendations, or similarity matching. In this post, we’ll walk through how to set up OpenAI embeddings inside an Ash resource.

We start by wrapping the OpenAI embeddings API in an Ash-compatible module:

# lib/my_app/open_ai_embedding_model.ex
defmodule MyApp.OpenAiEmbeddingModel do
  use AshAi.EmbeddingModel

  @impl true
  def dimensions(_opts), do: 3072

  @impl true
  def generate(texts, _opts) do
    api_key = System.fetch_env!("OPENAI_API_KEY")

    headers = [
      {"Authorization", "Bearer #{api_key}"},
      {"Content-Type", "application/json"}
    ]

    body = %{
      "input" => texts,
      "model" => "text-embedding-3-large"
    }

    response =
      Req.post!("https://api.openai.com/v1/embeddings",
        json: body,
        headers: headers
      )

    case response.status do
      200 ->
        response.body["data"]
        |> Enum.map(fn %{"embedding" => embedding} -> embedding end)
        |> then(&{:ok, &1})

      _ ->
        {:error, response.body}
    end
  end
end

This tells Ash how to produce embeddings for a list of texts. We’re using OpenAI’s text-embedding-3-large model, which outputs vectors of size 3072.

PostgreSQL needs the vector extension to store embeddings efficiently:

# lib/my_app/repo.ex
defmodule MyApp.Repo do
  use AshPostgres.Repo,
    otp_app: :my_app

  @impl true
  def installed_extensions do
    ["ash-functions", "citext", "vector"] # enable vector support
  end

  @impl true
  def prefer_transaction?, do: false
  @impl true
  def min_pg_version, do: %Version{major: 16, minor: 0, patch: 0}
end

Let’s embed product descriptions:

# lib/my_app/shopping/product.ex
defmodule MyApp.Shopping.Product do
  use Ash.Resource,
    otp_app: :my_app,
    domain: MyApp.Shopping,
    extensions: [AshAi, AshOban],
    data_layer: AshPostgres.DataLayer

  vectorize do
    full_text do
      text fn product ->
        """
        Name: #{product.name}
        Description: #{product.description}
        """
      end
    end

    attributes description: :vectorized_description

    strategy :ash_oban
    ash_oban_trigger_name :my_vectorize_trigger

    embedding_model MyApp.OpenAiEmbeddingModel
  end

  oban do
    triggers do
      trigger :my_vectorize_trigger do
        action :ash_ai_update_embeddings
        worker_read_action :read
        worker_module_name __MODULE__.AshOban.Worker.UpdateEmbeddings
        scheduler_module_name __MODULE__.AshOban.Scheduler.UpdateEmbeddings
        queue :default
      end
    end
  end

  postgres do
    table "products"
    repo MyApp.Repo
  end

  actions do
    defaults [:read, :create, :update, :destroy]
    default_accept [:name, :description]
  end

  attributes do
    uuid_v7_primary_key :id
    attribute :name, :string, public?: true
    attribute :description, :string, public?: true
    timestamps()
  end
end

This setup ensures that whenever a product is created or updated, its description is vectorized in the background using Oban.

Ash will generate the necessary migrations to support vector columns:

mix ash_postgres.generate_migrations add_vector_support
mix ash_postgres.migrate

With just a few modules and configuration changes, you now have OpenAI embeddings stored directly in PostgreSQL, ready for similarity queries. Combined with Ash’s vector features, you can build powerful semantic search and recommendation systems on top of your data with minimal effort.