living-memory

Agent memory for any AI agent, as plain tables that any tool, model or
person can read and any system can share. living-memory keeps the full
conversation, normalizes each JSON value into related tables (Codd's normal
form) written as Multi-Sheet Tab-Separated Values (MTSV), stores them in a
data bank that only ever adds, and gives back what is new. Any JSON source
comes in through its own integration; Claude Code is the first.

Why It Is Useful

AI agents need memory that lasts, and teams need to share it. Summaries lose
detail, raw logs repeat on every turn and lock to one tool's format, and a
database behind one app is closed to the next agent. MTSV tables keep each
fact once, open in spreadsheets, grep, dataframes, SQL and language models,
and convert to Excel (XLSX), ODS, CSV, JSON, SQLite, Parquet and Arrow with
mtsv.

Get Started

• Install: pipx install living-memory
• Give Claude Code memory: living-memory --install=claude-code, which states the change it makes and asks first
• Convert JSON in Python: living_memory.convert(values, schema), with the JSON Schema that describes the values
• The library, the command and each integration: python/README.md

Details

• Specification: spec/living-memory.mtsv
• Format: MTSV, an IETF Internet-Draft
• Requirements: Python 3.11 or later; the command runs on Linux and macOS, the library on any platform; mtsv, installed with it
• Version: the version field of python/pyproject.toml; each release tagged vX.Y.Z and published on PyPI
• Layout: spec/, the specification; conformance/, the tests every implementation passes; python/, the Python implementation
• License: MIT

Get Help

• Issues: living-memory issue tracker
• Security: SECURITY.md
• Maintainer: Demos Ra
• More: all MTSV tools by Demos Ra