ReqLLM • OpenAI Quickstart (gpt-4o-mini)

Minimal, practical examples for text generation, streaming, and structured outputs using the inexpensive OpenAI gpt-4o-mini model.

Tip: In Livebook, add a secret named OPENAI_API_KEY (preferred), or paste your key in the setup cell below.

Install

Mix.install([
  {:req_llm, "~> 1.0.0-rc.3"}
])

Configure OpenAI key

# Uses OPENAI_API_KEY from env/secrets if present, otherwise prompts you to paste one.
model = "openai:gpt-4o-mini"

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

if key in [nil, "", "sk-..."] do
  raise """
  Missing OpenAI key.
  - Add a Livebook secret named OPENAI_API_KEY, or
  - Replace the placeholder in this cell with your real key.
  """
else
  ReqLLM.put_key(:openai_api_key, key)
  :ok
end

1) Basic text generation

Change the prompt and re-run.

prompt = "In one sentence, explain what Elixir Livebook is."

text = ReqLLM.generate_text!(model, prompt)

text

2) Streaming tokens as they arrive

ReqLLM.stream_text!(model, "Write a 3–4 sentence upbeat intro to functional programming.")
|> Enum.each(&IO.write/1)

IO.puts("") # newline at the end

3) Structured data (JSON-like object)

Provide a simple schema; the model returns a validated Elixir map.

schema = [
  idea: [type: :string, required: true],
  steps: [type: {:list, :string}, required: true]
]

object =
  ReqLLM.generate_object!(
    model,
    "Return a very short DIY idea and 3 concise steps as JSON matching the schema.",
    schema
  )

object

That’s it—three essentials to get you productive:

• generate_text!/3 for simple output
• stream_text!/3 for streaming
• generate_object!/4 for structured results