03 / 06
Open source
Memory Vault AI
A model-agnostic memory layer that gives any LLM persistent, user-scoped context, shipped as a Python package, REST API, CLI and MCP server.
- Role
- Solo, library, API, CLI and docs
- Year
- 2026
- Stack
- Python · FastAPI · ChromaDB · MCP · Docker · PyPI
- Reading time
- Project note
- Team
- Solo project

01 · Overview
What it is
A model-agnostic memory layer that gives any LLM persistent, user-scoped context, shipped as a Python package, REST API, CLI and MCP server.
Memory Vault sits between an application and any LLM: it classifies what is worth remembering, compresses what is stale, and injects only the relevant memories into the next prompt. It runs locally against ChromaDB, so the default deployment costs nothing and never sends memory to a vendor.
02 · Highlights
Highlights
- 01
Four memory types (episodic, semantic, working, procedural) with vector retrieval.
- 02
Token-budget-aware recall; old memories are compressed instead of deleted.
- 03
Local-first: ChromaDB and local embeddings, no vendor required.
- 04
MCP-compatible with Claude Desktop and Cursor.
- 05
Published on PyPI as `memory-vault`.
03 · Stack
Stack
- Python
- FastAPI
- ChromaDB
- MCP
- Docker
- PyPI
Sources
- Memory Vault AI - repository READMEOpen (opens in a new tab)
- PyPI - memory-vaultOpen (opens in a new tab)
- GitHub - /users/zidanmubarak/reposOpen (opens in a new tab)