Agent with memory using Microsoft Agent Framework

This example demonstrates how to add memory to a Microsoft Agent Framework agent using a custom ContextProvider, i.e., a UserInfoMemory as follows:

• invoked: runs after each agent call and extracts name/age from user messages.
• invoking: runs before each agent call and injects memory into the prompt as instructions.
• serialize: returns JSON so memory can be persisted with the thread/session.

Requirements

• Python 3.10+
• An OpenAI API key set as an environment variable (OPENAI_API_KEY)

Steps for running this example in the shell

1. Install dependencies:

python -m venv .venv

# macOS/Linux:
source .venv/bin/activate

# Windows Command Prompt:
.venv\Scripts\activate.bat

# Windows PowerShell:
.venv\Scripts\Activate.ps1

pip install -r requirements.txt

2. Export your API key as an environment variable:

export OPENAI_API_KEY="sk-..." # Windows cmd: set OPENAI_API_KEY="sk-..." # Windows PowerShell: $env:OPENAI_API_KEY="sk-..."

3. Run the script:

python agent_with_memory.py

Output

The script will demonstrate the agent's ability to remember user facts:

Agent ready. Type 'exit' to quit.

Try asking a question first — it should request name/age.

You: What is 2+2?

Agent: I’d be happy to help with that, but first, could you please tell me your name?

You: John

Agent: Nice to meet you, John! Can you also tell me your age?

You: 25

Agent: Thank you for sharing that, John! Now, to answer your question, 2 + 2 equals 4. If you have any more questions, feel free to ask!