OpenAI

OpenAI will return high-quality extraction for little money and usually a lot faster than a local Ollama setup. If you run the process in your terminal the code will look for the usual OPENAI_API_KEY.

Linux / macOS

export OPENAI_API_KEY=...

Windows (PowerShell)

$env:OPENAI_API_KEY = "..."

Windows (cmd.exe)

setx OPENAI_API_KEY "..."

or in the code (look for os.environ.get("OPENAI_API_KEY", "")).

There are two OpenaAI model parameters you can set:

• --extraction-model: the model used for summary, title and graph extraction. Default is gpt-4o-mini.
• --discovery-model: used for language and schema detection. Default is gpt-4o.

These are used only if you enable the OpenAI flag, like so:

knwler extract --file <file path> --openai --extraction-model gpt-5.2

See also the batch processing info to process directories using OpenAI’s batch API.