Memory Systems

Hands-on lab with starter code you can copy + run, and guided steps to produce the artifact for this module.

Setup

Open a fresh terminal in a new folder, e.g. mkdir lab-3 && cd lab-3.
Create a Python 3.11 venv: python -m venv .venv && source .venv/bin/activate.
Install the dependencies noted on each starter snippet (the small pip install … line above the code).
Copy each starter into a file with the suggested name, then run it and capture the output.

Starter code

01 · Long-term memory with pgvector

pip install psycopg[binary] openai

save as: 03_01_lab.py

python

import psycopg, openai
db = psycopg.connect("postgresql://localhost/agent")
db.execute("CREATE EXTENSION IF NOT EXISTS vector;")
db.execute("""CREATE TABLE IF NOT EXISTS mem(
    id bigserial primary key, text text, embedding vector(1536))""")

def embed(text):
    return openai.embeddings.create(model="text-embedding-3-small", input=text).data[0].embedding

def remember(text):
    db.execute("INSERT INTO mem(text, embedding) VALUES(%s, %s)", (text, embed(text)))
    db.commit()

def recall(query, k=3):
    return db.execute(
        "SELECT text FROM mem ORDER BY embedding <-> %s LIMIT %s",
        (embed(query), k)).fetchall()

Run with: python 03_01_lab.py

02 · Sliding-window + summary short-term memory

save as: 03_02_lab.py

python

class ChatMemory:
    def __init__(self, llm, window=8):
        self.llm, self.window, self.msgs, self.summary = llm, window, [], ""
    def add(self, role, content):
        self.msgs.append({"role": role, "content": content})
        if len(self.msgs) > self.window:
            chunk = self.msgs[:-self.window]
            self.summary = self.llm.summarise(self.summary + "\n" + str(chunk))
            self.msgs = self.msgs[-self.window:]
    def context(self):
        return [{"role":"system","content": f"Summary: {self.summary}"}] + self.msgs

Run with: python 03_02_lab.py

Guided steps

Work through each step in order. Stop when your output matches the expected artifact — then push your work to your lab folder.

1
Pgvector memory
Do — Stand up Postgres+pgvector locally. Ingest 100 personal notes and query top-3 by embedding distance.
Expected artifact — Query 'what did I learn about LoRA' returns notes that actually mention LoRA in rank 1.
Run free:
2
Sliding window + summary
Do — Implement ChatMemory with window=6. Run a 20-turn conversation and dump the rolling summary.
Expected artifact — memory_dump.json with 6 recent messages and a summary field referencing earlier topics.
Run free:
3
Memory eval
Do — On turn 25, ask 'what was the user's name they gave on turn 2?'. Test with and without long-term memory.
Expected artifact — With memory: correct. Without: incorrect/unsure. Documented in memory_eval.md.
Run free:

Authoritative sources

Independent primary sources used to check and corroborate the guidance on this page.

NIST National Initiative for Cybersecurity Education (NICE)
US national framework for cybersecurity work roles and competencies.
Financial Action Task Force (FATF)
Global AML/CFT standards referenced by financial-crime curricula.
FBI Internet Crime Complaint Center — Annual Reports
Primary loss and typology data used in fraud-investigation training.
US Federal Trade Commission — Consumer Sentinel Data Book
Official complaint statistics for teaching fraud trend analysis.