AI Email Insight Agent

I receive roughly 100-200 messages every day and I’m looking for a small agent that can read an exported inbox file ( .eml, .mbox or even a simple CSV ) and return only the ten things I actually have to act on today.

The focus is strictly on my Business folder, and within those messages the agent must surface:
• clear action items first
• a concise summary of the day’s key emails
• any hard deadlines or meetings I should not miss

personal, property, promo mail will be secondary needs but needs to be included in insights as needed

Here’s what I need from you:

1. Code (Python preferred, open to Node.js) that:
– parses the chosen export format,
– uses an LLM or local model (OpenAI, GPT-4o, Llama-3 etc.) to classify and rank messages,
– outputs a single markdown or HTML report with the ten insights, each tagged as Action, Summary, Deadline or Meeting.

2. An explanation of the prompting / system instructions you use so I can refine them myself later. Please comment the code generously so I can tweak model parameters, temperature, and filtering logic without sharing credentials.

3. A short README covering:
– how to export email safely,
– how to run the script locally (virtual-env, required packages),
– how to swap in a different LLM or an offline model.

Acceptance criteria
• Running the script on a sample mbox generates a dated report featuring exactly ten bullet points, each no longer than 280 characters.
• Action items appear before summaries, deadlines, and meetings.
• No external email access is required; everything runs from the local export.

If you can think of smarter classification methods, feel free to suggest them—just keep the setup simple enough for a non-developer to repeat.