What You’ll Learn

• The difference between session memory and persistent memory
• How to decide what’s worth remembering across sessions
• How to implement a file-based memory store

The Problem

After a session ends, everything the agent learned is gone. Next session starts from zero. The agent re-learns preferences, re-discovers conventions, re-reads the same files.

The Solution

A memory system that persists cross-session knowledge. But with one rule: only persist what cannot be re-derived from current work.

Memory types:
+------------------+     +------------------+     +------------------+
| Project memory   |     | User memory      |     | Session memory   |
| (deps, patterns) |     | (preferences)    |     | (current state)  |
+------------------+     +------------------+     +------------------+

How It Works

1. A MemoryStore backed by JSON files on disk.

2. The agent can read_memory(key) and write_memory(key, value).

3. Rules prevent garbage: memory only for preferences, conventions, decisions — not for things that can be re-read from the filesystem.

class MemoryStore:
    def __init__(self, path: Path):
        self.path = path
        self.data = json.loads(path.read_text()) if path.exists() else {}

    def get(self, key: str) -> str:
        return self.data.get(key, "")

    def set(self, key: str, value: str):
        self.data[key] = value
        self.path.write_text(json.dumps(self.data, indent=2))

What Changed From s08

Component	Before (s08)	After (s09)
Cross-session	Nothing	Persistent memory store
Preferences	Re-learned	Remembered across runs
Decisions	Lost	Archived as memory

Try It

cd learn-claude-code
python agents/s09_memory_system.py

1. Set a memory: this project uses pytest for testing
2. Set a memory: I prefer snake_case over camelCase
3. Exit, restart, and What do you remember about this project?

Key Takeaway

Memory gives direction; current observation gives truth. Only persist what cannot be re-derived from current work.