Memory Types & Concepts
Rekall models AI memory across seven distinct types, each optimized for different kinds of knowledge. Understanding these types is key to building agents that remember, learn, and improve over time.
Getting started?
If you are new to Rekall, start with Episodic Memory and Semantic Memory. These two types cover the most common agent memory needs.
Memory Types
Episodic
Event-based memories capturing what happened, when, where, and who was involved. Ideal for conversation history, meeting notes, and significant interactions.
Semantic
Knowledge graph with entities and typed relationships. Stores facts, skills, preferences, and connections between concepts.
Procedural
Detected workflows and step-by-step procedures. Automatically learned from repeated patterns or explicitly defined.
Long-Term
Consolidated knowledge extracted from episodic memories. Features natural decay based on access frequency and importance.
Short-Term
Redis-backed session context with automatic TTL expiration. Fast read/write for current conversation state.
Execution
Agent task state management with checkpoint, pause, and resume. Enables long-running workflows that survive restarts.
Preferences
Intelligent preference learning from user interactions. Tracks patterns in choices, coding style, and communication preferences.
Comparison Table
| Type | Storage | TTL / Lifecycle | Use Case |
|---|---|---|---|
| Episodic | PostgreSQL + Qdrant | Decays over time | Conversation history, events, meeting notes |
| Semantic | Neo4j + Graphiti | Persistent | Knowledge graphs, entity relationships, facts |
| Procedural | PostgreSQL | Persistent | Workflows, checklists, learned processes |
| Long-Term | PostgreSQL + Qdrant | Importance-based decay | Consolidated knowledge, patterns, insights |
| Short-Term | Redis | TTL (default 3h) | Session context, working state, temp data |
| Execution | PostgreSQL | Task lifetime | Agent task state, checkpoints, pause/resume |
| Preferences | PostgreSQL | Persistent | User preferences, style patterns, choices |
Memory Contexts
Memories exist within contexts that control isolation and sharing. Rekall supports three context types:
Decay & Consolidation
Rekall models memory strength using an Ebbinghaus-inspired decay curve. Memories naturally fade over time unless reinforced through access. The consolidation process periodically extracts patterns from episodic memories into long-term storage, preserving key insights while allowing details to decay.
Learn about decay and consolidation →Relationships
Memories do not exist in isolation. Rekall tracks cross-memory connections including semantic relationships between entities, references from episodic memories to entities, and temporal chains of events.
Learn about memory relationships →