# Memorg

> Memorg is an external memory layer for LLM apps. It is a pip-installable Python library (Python 3.11+, MIT-licensed) that combines hierarchical SQLite storage, a USearch vector index, OpenAI embeddings, and an MCP server (FastMCP). The memory model is four-level: session → conversation → topic → exchange, plus typed memory items (documents, notes, anything). Retrieval blends semantic similarity, recency, and importance, and trims results to a per-session token budget.

Memorg is designed for app builders shipping LLM chat where conversation continuity matters, and for RAG teams who outgrew naive vector search. The library runs in your process against a single SQLite file. There is no managed service required.

Memorg is a Neul Labs project (https://www.neullabs.com).

## Product

- [Home](https://memorg.neullabs.com/): What Memorg is, what it ships with, and the pitch
- [Features](https://memorg.neullabs.com/features/): The full feature set — hierarchy, semantic search, blended ranking, token budget, single-file SQLite, MCP
- [How it works](https://memorg.neullabs.com/how-it-works/): Architecture end to end — hierarchy, storage, vectors, and the retrieval path
- [Quickstart](https://memorg.neullabs.com/quickstart/): Install with pip, create a session, store exchanges, call search_context()
- [Use cases](https://memorg.neullabs.com/use-cases/): Where Memorg fits — chat continuity, RAG upgrade, agent memory, support recall, document memory
- [FAQ](https://memorg.neullabs.com/faq/): Answers grouped by basics, retrieval, operations, and integration
- [Glossary](https://memorg.neullabs.com/glossary/): Definitions of the core concepts
- [About](https://memorg.neullabs.com/about/): Design thesis — why memory should be infrastructure, not a framework
- [Blog](https://memorg.neullabs.com/blog/): Engineering essays on memory design
- [Docs](https://docs.neullabs.com/memorg/): API reference and guides

## Use cases

- [Conversation continuity](https://memorg.neullabs.com/use-cases/chat-continuity/): Keep chat coherent across turns and sessions without re-sending the transcript
- [Upgrade naive RAG](https://memorg.neullabs.com/use-cases/rag-upgrade/): Move past a bare vector store to retrieval that understands users, recency, and importance
- [Agent memory](https://memorg.neullabs.com/use-cases/agent-memory/): Deterministic recall next to any orchestrator, without adopting a whole runtime
- [Support-bot recall](https://memorg.neullabs.com/use-cases/support-recall/): Per-customer memory so the bot resumes where the last conversation ended
- [Documents as memory](https://memorg.neullabs.com/use-cases/document-memory/): Store notes and documents alongside conversations and search across them

## Blog posts

- [Your chat app's memory is a 50-line bandage on RAG](https://memorg.neullabs.com/blog/chat-app-memory-50-line-bandage-on-rag/): Why a small dedicated memory layer beats the chat-history-table-plus-vector-store-plus-rolling-summary stack at the seam where it breaks
- [Forgetting as a feature: aging memory in LLM agents](https://memorg.neullabs.com/blog/forgetting-as-a-feature/): Why a good memory layer needs to forget on purpose, and how importance and recency scoring make forgetting tractable without explicit deletes
- [Designing memory for agents vs memory for chatbots](https://memorg.neullabs.com/blog/memory-for-agents-vs-memory-for-chatbots/): How read patterns, write rates, and scoping diverge between agents and chatbots, and why Memorg's hierarchy serves both

## Compare

- [vs Mem0](https://memorg.neullabs.com/compare/mem0/): Memorg is library-first and deterministic; Mem0 is a managed offering with LLM-extracted facts
- [vs Letta (MemGPT)](https://memorg.neullabs.com/compare/letta/): Letta is an agent runtime; Memorg is just the memory layer
- [vs Zep](https://memorg.neullabs.com/compare/zep/): Zep is a temporal knowledge-graph memory service with a cloud offering; Memorg is a local-first, single-file SQLite library with deterministic retrieval
- [All comparisons](https://memorg.neullabs.com/compare/): The comparison hub

## Optional

- [Full corpus](https://memorg.neullabs.com/llms-full.txt): Every key page concatenated for LLM ingestion
- [RSS](https://memorg.neullabs.com/rss.xml): Blog feed
- [GitHub](https://github.com/neul-labs/memorg): Source, issues, releases
- [Neul Labs](https://www.neullabs.com): The org behind Memorg