Questions teams ask before they pip install.

Memorg is an external memory layer for LLM apps — a pip-installable Python library that combines hierarchical SQLite storage, a USearch vector index, OpenAI embeddings, and an MCP server. It stores and retrieves context so LLM apps have longer, smarter conversations without forgetting.

No. It is a Python library that uses a vector index (USearch) and a relational store (SQLite) together. The vector index handles similarity search; SQLite holds the hierarchy, metadata, and source-of-truth content. You get vector recall without running a vector database.

No. Memorg is deliberately just the memory layer. It does not own your prompts, run your agent loop, or ship a router, planner, or evaluator. You call it from any handler or orchestrator you already use.

A blended, deterministic one. Queries are embedded with OpenAI, USearch returns nearest neighbours, and results are re-scored with recency and importance before being trimmed to the session token budget. There is no LLM reranker in the default path — the scoring is deterministic.

A per-session limit on how much retrieved context comes back. Memorg trims results to fit the budget you set when you create the session, so the returned payload always fits the model window.

Sessions hold conversations, conversations hold topics, topics hold exchanges. That lets you scope a query to a single topic, a whole conversation, or a user’s entire history — instead of searching one flat index.

Storage and vector search are local. Embedding and generation call OpenAI, which requires a network and an OPENAI_API_KEY. Swapping in another embedder is a contained change because the vector-store interface is small.

USearch keeps the index small and fast even at millions of items, and SQLite handles the metadata fine at that scale. The practical limit is your embedding budget and how aggressively you prune low-importance items.

pip install memorg, Python 3.11+. Dependencies include openai, tiktoken, aiosqlite, numpy, usearch, and fastmcp. There are no native services to run.

The package is published on PyPI and the project is in alpha (per its own classifier). The data model, retrieval API, and MCP server are stable enough to build on; expect refinements at the edges.

Memorg ships an OpenAI client integration for embeddings and chat. The bundled FastMCP server exposes Memorg to any MCP-aware client (Claude Desktop, Cursor, and others). For other providers, use Memorg as a library and feed the retrieved context into whatever generation call you already make.

Those frameworks include memory as one feature among many. Memorg is just the memory layer — a small surface you can drop next to any orchestrator. It does not own your prompts, tools, or agent loop. See the Mem0, Letta, and Zep comparisons for direct memory-tool trade-offs.

If you are running one-off completions with no continuity, you do not need a memory layer. If you need a hosted service with multi-tenant isolation and a control plane, or a temporal knowledge graph as the primary query shape, another tool may fit better.