memory-core-v2

name	memory-core-v2
description	Use when integrating Agent Memory Core v2 with an LLM agent or reviewing its deterministic memory gates.
version	2.0.0-dev
author	xPerryx + Lena OpenClaw
license	MIT
platforms	["linux","macos","windows"]

Memory Core v2

Local-first, framework-agnostic long-term memory for LLM agents (Hermes, OpenClaw, and compatible hosts). A strict deterministic gate pipeline prevents LLM hallucinations, prompt injections, and schema violations from entering the memory store.

Features

• G1–G9 Gate Pipeline — pure-functional, fail-closed, LLM-free validation
• Human-in-the-Loop Queue — privileged lanes route to a proposal_queue for human approval; approvals atomically commit facts
• Append-only Audit Log — every decision recorded with a strict reason code
• Asynchronous Consolidator — background digestion of episodes into facts
• Namespace Isolation — per-namespace or single-file SQLite topologies
• Zero Dependencies — core pipeline uses only the Python Standard Library
• CLI (memory-cli) — init, db path, queue ls/approve/reject, audit, run

Status

• B1: Skeleton, config, schema, storage, episodes, migration
• B2: Gate pipeline (G1–G9), proposal queue, audit
• B3: Consolidator run loop, LLM backends
• B4: Narrative layer, injection, adapters
• B5: CLI, packaging, docs

Quick Reference

# Initialize
memory-cli init my_agent

# Run consolidator
OPENAI_API_KEY="sk-..." memory-cli run my_agent

# Review proposals
memory-cli queue ls my_agent
memory-cli queue approve my_agent <id> --by admin

# Audit trail
memory-cli audit my_agent -n 20

See the master specification for full documentation.