HyperLLM

Call all LLM APIs using the OpenAI format.

Installation

Add hyper_llm to your list of dependencies in mix.exs:

def deps do
  [
    {:hyper_llm, "~> 0.6.0"}
  ]
end

Configurations

config :hyper_llm,
  openai: [
    api_key: "sk-..."
  ],
  anthropic: [
    api_key: "sk-..."
  ],
  # ...

Usage

The HyperLLM.Conversation module is used to start a new conversation of the LLM provider of your choices.

HyperLLM.Conversation.start(model: "openai/gpt-4o-mini")
|> HyperLLM.Conversation.append(:developer, "You are a helpful assistant.")
|> HyperLLM.Conversation.append("Spell strawbeerry")
|> HyperLLM.Conversation.run!()
#=> %HyperLLM.Conversation{
  model: "openai/gpt-4o-mini",
  thread: [
    %{role: "developer", content: "You are a helpful assistant."},
    %{role: "user", content: "Spell strawbeerry"},
    %{role: "assistant", content: "Strawberry. 🍓"}
  ]
}

Under the hood, the HyperLLM.Conversation module is using HyperLLM.Chat.completion/3 to receive a OpenAI compatible response.

HyperLLM.Chat.completion("openai/gpt-4o-mini", [
  %{role: "user", content: "Spell \"strawberry\""}
])
#=> {:ok, %{
  "id" => "chatcmpl-1234567890",
  "object" => "chat.completion",
  "created" => 1717753987,
  "choices" => [
    %{
      "message" => %{
        "role" => "assistant",
        "content" => "Strawberry. 🍓"
      },
      "index" => 0,
      "finish_reason" => "stop"
    }
  ]
}}

HyperLLM.Chat.completion("anthropic/claude-3-5-sonnet-20240620", [
  %{role: "user", content: "Spell \"strawberry\""}
])
#=> {:ok, %{
  "id" => "chatcmpl-1234567890",
  "object" => "chat.completion",
  "created" => 1717753987,
  "choices" => [
    %{
      "message" => %{
        "role" => "assistant",
        "content" => "Strawberry. 🍓"
      },
      "index" => 0,
      "finish_reason" => "stop"
    }
  ]
}}

If you are using Phoenix, you can use the HyperLLM.Conversation module in your LiveView.

defmodule ChatLive do
  use Phoenix.LiveView

  def render(assigns) do
    ~H"""
    <div>
      <dl>
        <%= for message <- @chat.messages do %>
          <dt><%= message.role %></dt>
          <dd><%= message.content %></dd>
        <% end %>
      </dl>
    </div>
    """
  end

  def mount(params, session, socket) do
    {:ok,
    socket
    |> assign(conv: HyperLLM.Conversation.start(model: "openai/gpt-4o-mini"))}
  end

  def handle_event("send_message", %{"message" => message}, socket) do
    conv = HyperLLM.Conversation.append(socket.assigns.conv, :user, message)

    send(self(), :chat_completion)
    
    {:noreply, socket |> assign(conv: conv)}
  end

  def handle_info(:chat_completion, socket) do
    with {:ok, conv} <- HyperLLM.Conversation.run(socket.assigns.conv) do
      {:noreply, socket |> assign(conv: conv)}
    end
  end
end

Provider Support

Provider	Completion	Streaming
Anthropic	✅	❌
Cloudflare	✅	❌
Groq	✅	❌
LlamaCPP	✅	❌
Mistral	✅	❌
Ollama	✅	❌
OpenAI	✅	❌
vLLM	✅	❌
xAI	✅	❌
Azure	❌	❌
AWS SageMaker	❌	❌
AWS Bedrock	❌	❌
Cohere	❌	❌
DeepSeek	❌	❌
Empower	❌	❌
Google - Vertex AI	❌	❌
Google - Palm	❌	❌
Google AI Studio	❌	❌
Hugging Face	❌	❌
Perplexity	❌	❌
Replicate	❌	❌
TogetherAI	❌	❌
Vertex AI	❌	❌