MetaMemory vs Letta (MemGPT)

TL;DR

• Letta (formerly MemGPT) pioneered LLM self-editing memory, where the language model manages what gets stored and retrieved.
• Letta is a full agent framework with built-in memory; MetaMemory is a dedicated memory engine that integrates with any framework.
• MetaMemory uses 4 embedding types and 5 retrieval channels with Reciprocal Rank Fusion for more granular and adaptive memory retrieval.
• Letta's core memory blocks offer a novel approach to tiered memory; MetaMemory's multi-vector approach provides finer-grained memory categorization and adaptive learning.

Feature Comparison

Architecture Comparison

The core distinction is scope. Letta is a complete agent framework where memory management is one component of a larger system. MetaMemory is a purpose-built memory engine designed to integrate with any agent framework via API.

Memory Architecture

Letta introduced a novel tiered memory architecture inspired by operating system memory management. Core memory blocks (such as human and persona blocks) act as a working memory that the LLM can read and edit directly. Archival memory serves as long-term storage with vector search retrieval, and recall memory provides searchable conversation history. This LLM-driven self-editing approach lets the agent decide what information is important enough to persist.

MetaMemory takes a different approach with four distinct embedding types: semantic (factual knowledge), emotional (sentiment and affect), process (how-to instructions), and context (events with temporal metadata). Each type is stored in its own vector space with dedicated retrieval channels, providing more granular memory categorization without relying on the LLM to make storage decisions.

Self-Editing Memory vs Conflict Resolution

Letta's pioneering contribution is LLM-driven memory editing. The language model itself decides when to update, add, or remove memories from core and archival storage. This creates a flexible system where memory management emerges from the LLM's reasoning capabilities.

MetaMemory handles memory updates through a dedicated conflict resolution system that detects contradictory information and resolves it based on recency, confidence scores, and source reliability. This deterministic approach provides more predictable memory behavior without consuming LLM tokens for memory management operations.

Framework vs Engine

Letta provides a complete agent framework including tool use, multi-step reasoning, and memory management in a single package. This is ideal for teams that want an end-to-end solution and are willing to build within Letta's framework. Letta is open source under the Apache 2.0 license, offering full transparency and the ability to self-host.

MetaMemory is an API-first memory engine that integrates with any agent framework, whether that is LangChain, CrewAI, AutoGen, or a custom stack. This makes MetaMemory the more flexible choice for teams that already have an agent architecture and need to add a dedicated memory layer without adopting a new framework.

When to Choose Which

Choose Letta if you want a complete, open-source agent framework with built-in memory management and the flexibility of LLM-driven self-editing memory. Letta is a strong choice for new projects that don't have an existing agent stack and want an integrated solution with a novel approach to tiered memory.

Choose MetaMemory if you want a dedicated memory engine for your existing agent stack. MetaMemory is the stronger choice for teams that need multi-vector embeddings across semantic, emotional, process, and emotional types, adaptive retrieval that improves over time, emotional intelligence, or multi-channel retrieval for higher recall on complex queries, all without adopting a new agent framework.