[P] [D] Self hosting an open source model or using an API like chatgpt

What is the better approach to develop small scale LLM application,

The usecase is: Feeding some data (possibly containing PII) and use LLM to extract certain information from documents

The problem is if I use third party API such as OpenAI then i would be providing personal information which I do not have any way of masking before hand

The other option is to run llm model on a server, but the model sizes are quite large and would cost …

api application chatgpt data documents extract hosting information llm machinelearning openai open source open source model pii scale small