Skip to main content

Supply API keys

Cloud providers (Gemini, OpenAI) need a credential; local models via Ollama need none. LangExtract reads credentials from environment variables, or you can pass one directly. This guide covers each option; for the exact per-provider precedence, see the API reference §4.

LANGEXTRACT_API_KEY (any cloud provider)​

The general-purpose variable. Set it and LangExtract uses it for whichever cloud provider your model_id selects.

export LANGEXTRACT_API_KEY="your-api-key"

Provider-specific variables​

Each cloud provider also reads its own variable, which takes precedence over LANGEXTRACT_API_KEY:

  • Gemini: GEMINI_API_KEY, then LANGEXTRACT_API_KEY.
  • OpenAI: OPENAI_API_KEY, then LANGEXTRACT_API_KEY.

Use these when you work with more than one provider and want a distinct key for each.

The api_key parameter​

You can pass a key directly to lx.extract instead of using the environment. Passing a key directly is convenient for a quick test; prefer an environment variable for anything you share or deploy.

import langextract as lx

result = lx.extract(
    text_or_documents="Tom looked nervous before the interview.",
    prompt_description="Extract each person and the emotion they feel.",
    examples=examples,
    model_id="gemini-3.5-flash",
    api_key="your-api-key",
)

A .env file​

LangExtract reads keys from the environment; it doesn't load a .env file on its own. Keep your key in .env:

LANGEXTRACT_API_KEY=your-api-key

then load it into the environment before extracting, with python-dotenv:

from dotenv import load_dotenv

load_dotenv()  # sets LANGEXTRACT_API_KEY in the environment

import langextract as lx
# lx.extract(...) now picks up the key as usual

Add .env to your .gitignore so the key is never committed.

See also​