Minna Memory

Cross-session persistence for AI agents.

The Session Continuity Problem

Without Memory:
   Session 1          Session 2          Session 3
   ┌─────────┐       ┌─────────┐       ┌─────────┐
   │ Context │       │ ??????? │       │ ??????? │
   │ Built   │       │ Lost!   │       │ Lost!   │
   └─────────┘       └─────────┘       └─────────┘

With Minna Memory:
   Session 1          Session 2          Session 3
   ┌─────────┐       ┌─────────┐       ┌─────────┐
   │ Context │──────►│ Context │──────►│ Context │
   │ + Store │       │ + Recall│       │ + Recall│
   └─────────┘       └─────────┘       └─────────┘
        │                 ▲                 ▲
        └─────────────────┴─────────────────┘
                    minna.db (SQLite)

Entity-Attribute-Value Model

ENTITY (what)     ATTRIBUTE (aspect)    VALUE (data)
─────────────     ────────────────      ────────────
"8me"             "purpose"             "Loop toolkit"
"8me"             "tier1_status"        "COMPLETE"
"user"            "prefers"             "dark mode"

Basic Usage

from minna_memory import MinnaMemory

memory = MinnaMemory(".spine/minna.db")

# Store memory
memory.store(
    entity="my-project",
    attribute="status",
    value="In development",
    confidence=1.0
)

# Recall memories
memories = memory.recall(entity="my-project")

# Search across all memories
results = memory.search("development")

# Relationships
memory.relate(
    from_entity="user",
    to_entity="my-project",
    rel_type="works_on"
)

High Portability

Single SQLite file = complete state:

project/
└── .spine/
    └── minna.db    ← Contains ALL memory

Scenario	How Minna Helps
Move to new machine	Copy one SQLite file
Share with teammate	Send the .db file
Backup/restore	Standard file backup
Multi-project	Each project has own .db

Confidence Decay

Memories decay over time unless verified:

# Initial confidence: 1.0
# After 90 days without verification: ~0.5

memories = memory.recall(
    entity="my-project",
    include_decay=True,
    half_life_days=90
)

# Verify to reset decay
memory.verify(memory_id=123)

Use Cases

Progress Tracking

previous = memory.recall(entity=task.id, attribute="progress")
if previous:
    start_from = previous[0]["value"]  # Resume

Error Pattern Learning

similar = memory.search(error.message)
if similar:
    fix = similar[0]["value"]  # Apply known fix

User Preferences

verbosity = memory.set_preference("output", "verbosity")

Best Practices

DO

✅ Store semantic information
✅ Use consistent entity names
✅ Set appropriate confidence
✅ Verify memories that are still accurate

DON’T

❌ Store large binary data
❌ Store sensitive credentials
❌ Forget to clean up stale memories

Next Steps

See full documentation in 8me repository
Practice: Lab 12: Memory Integration

← Back to Concepts