TypeAgent AMP Demo - Email Thread Analysis

A demonstration of TypeAgent's AMP (Agent Memory and Planning) capabilities using an incident response email thread.

What This Demonstrates

TypeAgent's Structured-RAG approach to agent memory, showcasing:

• Intent Distillation: Parse natural language to extract structured actions/requests
• Action Tracking: Track what actions were performed by whom
• Memory Write-Back: Store actions as queryable structured memories
• Audit Trail: Query complete action history for accountability
• Memory-Driven Actions: Past actions inform future decisions
• Entity & Relationship Extraction: Extract people, roles, systems, and connections

Quick Start

Prerequisites

pip install typeagent

Setup

1. Set your OpenAI API key:

export OPENAI_API_KEY="your-api-key-here"

2. Open the notebook:

jupyter notebook email_typeagentdemo.ipynb

3. Run all cells in order

Files

• email_typeagentdemo.ipynb - Main demonstration notebook
• email_thread.eml - Email thread dataset used for the demo

Dataset

The demo uses an incident response email thread (email_thread.eml) that contains:

• 5 email messages from different team members
• Discussions about a data pipeline outage
• Actions taken to resolve the issue
• Team member roles and responsibilities

Features

The notebook demonstrates:

• Loading and parsing email threads
• Indexing messages with TypeAgent's Structured-RAG
• Querying extracted knowledge (actions, entities, relationships)
• Comparing TypeAgent's approach with limitations

TypeAgent

TypeAgent is Microsoft's framework for building agents with structured memory and planning capabilities.

License

MIT