Setup

%{ title: “Cookbook: Chat Response”, description: “Send a single chat request with Jido AI and extract the returned text, model, and usage fields.”, category: :docs, order: 110, doc_type: :cookbook, audience: :intermediate, tags: [:docs, :guides, :cookbook, :chat, :ai, :livebook], draft: false, legacy_paths: [“/docs/chat-response”, “/docs/cookbook/chat-response”], prerequisites: [“/docs/getting-started/first-llm-agent”], learning_outcomes: [

"Send a single chat request with Jido.AI.Actions.LLM.Chat",
"Extract the returned text, model, and usage fields"

], livebook: %{

runnable: true,
required_env_vars: ["OPENAI_API_KEY"],
requires_network: true,
setup_instructions: "Set OPENAI_API_KEY or LB_OPENAI_API_KEY before running the request cell."

} }

Setup

This recipe keeps the flow deliberately small: configure a provider key, send one chat request, and read the structured result. If you want a full multi-turn agent afterward, continue to Build an AI Chat Agent.

Mix.install([
  {:jido, "~> 2.0"},
  {:jido_ai, github: "agentjido/jido_ai", branch: "main"},
  {:req_llm, "~> 1.6"}
])

Logger.configure(level: :warning)

Configure the provider key the same way as Your first LLM agent. In Livebook, store it as a secret named OPENAI_API_KEY, which becomes LB_OPENAI_API_KEY inside the notebook.

openai_key = System.get_env("LB_OPENAI_API_KEY") || System.get_env("OPENAI_API_KEY")

configured? =
  if is_binary(openai_key) do
    ReqLLM.put_key(:openai_api_key, openai_key)
    true
  else
    IO.puts("Set OPENAI_API_KEY as a Livebook Secret or environment variable to run the request cell.")
    false
  end

Request

Jido.AI.Actions.LLM.Chat is the compact request-response API. Run it through Jido.Exec.run/2 with a prompt and an optional system prompt:

chat_result =
  if configured? do
    Jido.Exec.run(
      Jido.AI.Actions.LLM.Chat,
      %{
        model: :fast,
        prompt: "Write a two-sentence welcome for a new Jido user.",
        system_prompt: "You are a concise, friendly assistant.",
        temperature: 0.2,
        max_tokens: 120
      }
    )
  else
    {:skip, :no_openai_key}
  end

IO.inspect(chat_result, label: "Chat result")

The successful shape is {:ok, %{text: ..., model: ..., usage: ...}}. model: :fast resolves through jido_ai model aliases, so the same recipe works if you change the provider backing that alias later.

Response

Pattern match the response map to get the assistant text and any usage data you want to log or display:

chat_text =
  case chat_result do
    {:ok, %{text: text, model: model, usage: usage}} ->
      IO.puts("Model: #{model}")
      IO.inspect(usage, label: "Token usage")
      text

    {:error, reason} ->
      "Chat request failed: #{inspect(reason)}"

    {:skip, :no_openai_key} ->
      "Skipped request. Set OPENAI_API_KEY or LB_OPENAI_API_KEY to run it."
  end

IO.puts(chat_text)

This is the smallest useful chat recipe: one prompt in, one extracted text response out. When you need tools or multi-turn state, move on to Tool Response or the full AI Chat Agent guide.