Use Gemini with n8n implementation seems to default to OpenAI

[werther777]

Hi.
I'm using crawl4ai on a local docker + endpoint exposed.
n8n is hosted.
The call to the /crawl endpoint is successful, but I receive this:

[
  {
    "status": "failed",
    "created_at": 1750755035.9841993,
    "error": "API token must be provided for LLMExtractionStrategy. Update the config.py or set OPENAI_API_KEY environment variable."
  }
]

This is the call:

{
  "urls": "https://some_url",
  "crawler_params": {
    "headless": true
  },
  "extraction_config": {
    "type": "llm",
    "params": {
      "llm_config": {
        "provider": "gemini/gemini-2.0-flash"
      },
      "schema": {
        "title": "RecipeUrlsToScrape",
        "type": "object",
        "properties": {
          "recipe_name": {
            "title": "Recipe Name",
            "description": "Name of the recipe",
            "type": "string"
          },
          "recipe_url": {
            "title": "Recipe Url",
            "description": "Recipe Url",
            "type": "string"
          }
        },
        "required": ["recipe_name", "recipe_url"]
      },
      "extraction_type": "schema",
      "instruction": "From the crawled content, extract all recipe names and url. Ensure the url is for a recipe and not any other content"
    }
  },
  "cache_mode": "bypass"
}

my docker .env contains the key:

CRAWL4AI_API_TOKEN=my_key
GEMINI_API_KEY=my_gemini_api_key
OPENAI_API_KEY=

Open API key is empty because I'm not using it. However, the implementation of /crawl seems to assume I am and it's trying to find it.

Any idea? Am I doing something wrong?

[hafezparast]

The issue is the request format — the /crawl endpoint doesn't recognize extraction_config as a top-level field, so your LLM config is being silently ignored and it falls back to defaults.

You have two options:

Option A: Use /llm/job (recommended for LLM extraction)

POST /llm/job
{
  "url": "https://some_url",
  "q": "From the crawled content, extract all recipe names and url. Ensure the url is for a recipe and not any other content",
  "provider": "gemini/gemini-2.0-flash",
  "schema": "{\"title\": \"RecipeUrlsToScrape\", \"type\": \"object\", \"properties\": {\"recipe_name\": {\"type\": \"string\"}, \"recipe_url\": {\"type\": \"string\"}}, \"required\": [\"recipe_name\", \"recipe_url\"]}"
}

Option B: Use /crawl with extraction_strategy inside crawler_config

POST /crawl
{
  "urls": ["https://some_url"],
  "crawler_config": {
    "extraction_strategy": {
      "type": "LLMExtractionStrategy",
      "llm_config": {
        "provider": "gemini/gemini-2.0-flash",
        "api_token": "your-gemini-key"
      },
      "instruction": "From the crawled content, extract all recipe names and url...",
      "schema": { }
    },
    "cache_mode": "bypass"
  }
}

Also make sure your GEMINI_API_KEY env var is set in the Docker container (not just in .env — verify it's passed through in docker-compose.yml).