Mulberry (Very Alpha)

An AI package for Elixir that focuses on higher level application concerns. It will include an AI UI component library for Phoenix.

Installation

In your mix.exs:

While in active Alpha development it is probably best to pull it from Github

def deps do
  [
    {:mulberry, github: "agoodway/mulberry", branch: "main"}
  ]
end

def deps do
  [
    {:mulberry, "~> 0.1.0"}
  ]
end

Configuration

In your runtime.exs:

config :langchain, openai_key: System.get_env("OPENAI_API_KEY")
config :mulberry, brave_api_key: System.get_env("BRAVE_API_KEY")

Examples

Extract text from a PDF or image file in a Phoenix upload:

import Flamel.Wrap, only: [ok: 1]
alias Mulberry.Document
alias Flamel.Result

defp handle_upload(socket, entry) do
  {:ok, text} =
    consume_uploaded_entry(socket, entry, fn %{path: path} ->
      %{path: path, mime: entry.client_type}
      |> Document.File.new()
      |> Document.load()
      |> Result.map(fn file -> file.contents end)
      |> ok()
    end)

  socket
  |> assign(:file, AsyncResult.loading())
  |> start_async(:file, fn -> Files.parse_text(text) end)
end

The above example uses Flamel to handle :ok and :error tuples. See Result.map/2 and ok/1

Generate a summary of a web pages from a Brave Search API query:

alias Mulberry.Search.Brave
alias Mulberry.Document

# this chain can be applied to any Mulberry.Document whether it is a web page or an image file
def load_and_generate_summary(chain) do
  chain
  |> Chain.apply(&Document.load/1)
  |> Chain.apply(&Document.generate_summary/1)
  |> Chain.to_tuple()
end

Mulberry.search(Brave, "what is the best cake recipe?")
# apply this sequence of operations to each document returned from the search query
|> Enum.map(Chain.curry(&load_and_generate_summary/1))
# remove any results that resulted in an error
|> Enum.filter(&Result.ok?/1)

The chain above could be applied to ei

The above example uses Flamel chain a couple operations on the document and filter non-ok results. See Flamel.Chain and Result.ok?/1

TODO

• Text to audio
• Audio to text
• Video to text
• Add embeddings to Text module
• Typesense Search Adapter
• Streaming audio and video support
• Agents
• Implement other file types
• Backfill tests
• and much much more...

Documentation can be generated with ExDoc and published on HexDocs. Once published, the docs can be found at https://hexdocs.pm/mulberry.