6-layer SQLite-based memory with importance scoring, accessed via MCP server
| Layer | Purpose | Decay |
|---|---|---|
| Working | Short-term scratch space | Fast |
| Episodic | Specific events, conversations | Medium |
| Semantic | Facts, knowledge, concepts | Slow |
| Procedural | How-to patterns, behaviors | Very Slow |
| Meta | Self-reflection, insights | Very Slow |
| Identity | Identity, protected memories | Protected |
remember(content, layer?, metadata?)
recall(query, layer?, limit?)
query_layer(layer, options?)
save_to_layer(layer, content, metadata)
get_status()
High-value memories persist; mark what matters with importance scores
Content analyzed and placed in appropriate layer automatically
Memories linked: supports, contradicts, derived_from
Affective intensity stored for emotionally-weighted recall
Low-importance memories compressed, never deleted
Enterprise acceleration available (hardware dependent)
Docker containers. Production-ready. No configuration hell.
We'll configure CASCADE for your specific requirements.
Tell us what you're building. We'll tell you what you need.
We configure the system for your specific requirements.
Your AI has memory. Production-ready, running, yours.
CASCADE gives your AI system memory that actually works like memory should - with layers, decay, and persistence.
Start Your Project →