marinac-dev / openai

Elixir OpenAi Library with streaming support

OpenAi

Elixir OpenAi client library for with full support for streaming or SSE (Server Side Events).

Installation

Not available in Hex, but the package can be installed from git by adding openai to your list of dependencies in mix.exs:

def deps do
  [
    {:openai, git: "[email protected]:marinac-dev/openai.git", branch: "master"},
  ]
end

Documentation can be generated with ExDoc by running mix docs.

Configuration

config :openai,
  api_key: System.get_env("OPENAI_API_KEY"),
  organization_key: System.get_env("OPENAI_ORGANIZATION_KEY")

If you want to use retry mechanism, you can configure it like this:

config :openai,
  api_key: System.get_env("OPENAI_API_KEY"),
  organization_key: System.get_env("OPENAI_ORGANIZATION_KEY"),
  retry_config: %{
    retries: 5,
    delay: 200
  }

Usage

Once configured in your config.ex file, you can use the client to call the OpenAi API instantly.

prompt = %{model: "gpt-3.5-turbo", messages: [%{role: "user", content: "Hello!"}], stream: true}
OpenAi.chat_completion(prompt)
{:ok, "Hello! How may I assist you today?"}

openai's Issues

Mint Transport Error returning as tuple of tuple

Encountered this recently

  defp parse_response({:error, %Mint.TransportError{reason: :timeout}} = error) do
    Logger.error("OpenAi request timed out with error: #{inspect(error)}")
    {:error, error}
  end

Should be

  defp parse_response({:error, %Mint.TransportError{reason: :timeout} = error}) do
  Logger.error("OpenAi request timed out with error: #{inspect(error)}")
  {:error, error}
end

Implement retry mechanism when OpenAI API timeout or returns 4XX response

We need a retry mechanism that is configurable, like Tesla's Retry middleware.
To handle errors like timeout or when server had an error while processing your request.

Add support for OpenAI GPT Functions

OpenAI introduced GPT Functions recently.
Support for this is currently work in progress

Add response structure

I think it will make future changes to API more manageable and usage of responses more Elixir friendly.

100% API endpoints support

Make callback function optional in streaming request

Currently, stream() function for POST requests has a required callback function that handles responses.
However, there are situations where users might need to handle responses themselves with custom implementation, and would prefer to omit the built-in callback function.
Therefore, I would like to make the callback function as a parameter.