Documentation
Learn how to use Mneme to build persistent, learning memory for your AI workflows.
Getting Started
What is Mneme?
Mneme is a persistent memory system for AI assistants. It remembers context across conversations, learns from failures, and helps you avoid repeating mistakes. Named after the Greek muse of memory.
How to Use Mneme
Use Mneme through our chat interface at /chat. Start chatting naturally and Mneme will automatically identify important facts, preferences, patterns, and especially failures from your conversations.
Your First Conversation
Just start chatting naturally. Mneme will automatically identify important facts, preferences, patterns, and especially failures from your conversations. These become persistent memories that inform future sessions.
Memory System
L0: Active Memories
New memories start in L0. They're active and accessible but not yet validated. If unused, they naturally decay over time—like short-term memory.
L1: Validated Memories
When a memory proves useful (retrieved multiple times or manually promoted), it moves to L1. L1 memories don't decay—they're your long-term knowledge.
L2: WHY Documentation
Every memory can have L2 documentation explaining WHY it matters, what anti-patterns to avoid, and context for the future. This is where deep understanding lives.
Memory Types
Failure (Highest Priority)
What went wrong and why. Failure documentation is 3-10× more valuable than success documentation—it tells you what NOT to do.
Correction
Misunderstandings that were fixed. These help prevent the same confusion from happening again.
Pattern
Recurring approaches that work. Your validated workflows and methods.
Preference
How you like things done. Communication style, code formatting, tool choices.
Fact
Static context about you or your project. Names, configurations, dependencies.
Knowledge
Concepts and frameworks you're learning. Domain knowledge that grows over time.
Failures First
Why Failures Matter More
Success tells you what works in one case. Failure tells you what to avoid in all cases. A single anti-pattern can save hours of debugging across dozens of future sessions.
Anti-Pattern Capture
When you discover something that doesn't work, Mneme captures it with full context: what was tried, why it failed, and what to do instead.
Pattern Detection
If the same failure appears multiple times, Mneme automatically suggests promoting it to a validated anti-pattern in L1.
SCMS Architecture
What is SCMS?
Sparse Contextual Memory Scaffolding—a cognitive architecture for AI systems. It's the framework that powers Mneme's memory system.
Sparse Activation
Only retrieve memories that are actually relevant. No context pollution. High signal-to-noise ratio.
Self-Healing Cognition
When contradictions or errors are detected, the Integrity Cluster ensures definitions stay consistent. The system corrects itself.