Understanding Memphora
Learn the core concepts and architecture that power Memphora's persistent memory layer.
What is Memphora?
Memphora is a persistent memory layer for AI agents that enables them to remember past interactions, store user preferences, and maintain context across conversations. Unlike traditional stateless AI systems, Memphora provides long-term memory that persists between sessions.
Think of Memphora as a “memory bank” for your AI applications—it stores, retrieves, and manages memories intelligently, allowing your AI to build relationships and maintain context over time.
Semantic Search
Memphora uses advanced vector similarity search to find relevant memories based on semantic meaning, not just keyword matching. This means you can search for “user preferences” and find memories about “favorite programming languages” or “preferred communication style.”
How it works:
- Memories are converted to high-dimensional vectors (embeddings)
- Search queries are also converted to vectors
- Vector similarity search finds the most relevant memories
- Results are ranked by relevance, recency, and importance
3-Layer Caching System
Memphora's exclusive caching architecture ensures ultra-fast memory retrieval:
L1: Hot Cache
In-memory cache for frequently accessed memories. Instant retrieval.
L2: Warm Cache
Fast access to recently used memories. Low-latency retrieval.
L3: Persistent Cache
Persistent storage for all memories (ChromaDB + SQLite/PostgreSQL). Reliable long-term storage that survives restarts.
Context Compression
Memphora intelligently compresses context to reduce token usage while preserving meaning and relevance. This results in significant cost savings when using LLMs.
The compression system uses advanced AI models to summarize, deduplicate, and optimize memory context before sending it to your LLM, achieving significant token reduction without losing important information.
Memory Merging & Conflict Resolution
Memphora automatically detects and merges similar memories, resolves contradictions, and maintains consistency across your memory store.
Automatic Operations:
- Detect similar memories and merge them
- Identify contradictions and resolve conflicts
- Update outdated information
- Maintain memory consistency