v2.0 Updated February 19, 2026
Changelog

Lightning Memory

AI that never forgets. A multi-tier persistence architecture with per-project isolation, semantic vector search, and 3-way active-active replication across distributed nodes.

16+ Databases Live 4,264 Vector Embeddings 3-Node Replication

The Problem

Every AI Conversation Starts From Zero

When Claude, ChatGPT, or any LLM hits its context window limit, everything is lost. Your project context, preferences, decisions, history — gone. The next conversation has no idea what happened in the last one.

Enterprise teams spend weeks re-explaining context. Custom solutions are fragile, expensive, and still lose data when sessions crash. MCP gives AI tools — but not memory.

The Solution

Persistent, Searchable, Replicated Memory

Lightning Memory provides true continuity across sessions. Come back days, weeks, or months later — your AI knows exactly where you left off, what decisions were made, what the rules are, and what happened across every conversation. Not through fragile hacks, but through a purpose-built distributed architecture.

Three-Tier Architecture

Memory is organized into three tiers, each serving a different purpose. Together they ensure nothing is ever lost.

Tier 1: Per-Project Redis (Source of Truth)

Instant — Sub-Millisecond

Every project gets its own isolated Redis databases for memory (what happened) and conscience (what the AI should know and do). 16+ databases across dedicated port ranges, each replicated 3-way across Node 1, Node 2, and Node 3 using KeyDB active-active replication.

This is the source of truth. When the AI starts a session, it loads its project's conscience first, then recalls relevant memories. Fast, isolated, and always available.

Tier 2: Unimatrix (Semantic Vector Search)

Fast — Cross-Project Intelligence

Unimatrix is a Qdrant vector database that embeds every memory and conscience entry from every project into a searchable vector space. Currently 4,264 vectors using all-MiniLM-L6-v2 (384 dimensions, CPU-only).

Search by meaning, not keywords. Ask "What has User B been working on?" or "Proxmox backup procedures" and get ranked results across all projects simultaneously. Filter by project, owner, category, or designation.

Tier 3: Archive (Permanent Record)

Permanent — Full History

PostgreSQL backup database (ITSM system) plus full conversation transcripts captured by Lightning Continuum. Every AI conversation is archived as a searchable transcript. Nothing is ever deleted.

When you need to find something from 3 months ago, the archive has it. Full-text search, date filtering, and complete audit trail.

Unimatrix: Semantic Knowledge Search

What is Unimatrix?

Unimatrix is the vector knowledge layer that makes all memories searchable by meaning. Every memory and conscience entry from every project is embedded into a 384-dimensional vector and stored in Qdrant. This enables semantic search — find information by what it means, not what exact words were used.

Unimatrix 1 is the Schomp Technologies collective. Future customers get their own isolated Unimatrix (2, 3, etc.) with complete data separation and controlled cross-tenant access.

Multi-Tenant Design

Each tenant gets their own Unimatrix number with completely isolated vector collections. No data crosses boundaries unless an explicit Adjunct relationship is created — a controlled, auditable, tiered access link between tenants.

Primary Adjunct

Full read/write access. Account manager level. Can create memories, update conscience, manage agents in the target Unimatrix.

Secondary Adjunct

Read + scoped write. Project contributor level. Can read all, write only within assigned project scope.

Tertiary Adjunct

Read only. Observer level. Can search and read customer data for monitoring and reporting. No write access.

Designation System

Every entity in the system — humans, projects, AI agents — has a unique, human-readable designation that encodes their identity, group, and home collective.

User A, Designation 1 of 2, Unimatrix 1
Chloe Weaver, Reporter 3 of 4, Unimatrix 1
User B, Primary Adjunct of Unimatrix 2, Home: Unimatrix 1

Unimatrix 1 Taxonomy — 24 Designations

GroupCountMembers
Humans5User A (Project 1, Project 2), User B (Project 1, Project 2, Customer Tenant)
Projects4LMCP, ITSM, DB, Forge — each with own memory + conscience databases
Daily Planet10Clark Mercer, Lois Archer, Chloe Weaver, Jimmy Novak, Perry Ashton, Copy Editor, Archivist, Columnist, Copy Boy, Stringer
Infrastructure3Watchdog (health), Diagnostician (AI diagnosis), Taskmaster (builder)
Orchestration2Arbiter (session broker), Magistrate (task orchestrator)

Replication & Failover

Every memory database is replicated 3-way across all nodes using KeyDB active-active replication. If any node goes down, the others continue serving without data loss.

Node 1

Primary node. All 16+ Redis instances. Unimatrix vector DB. Full read/write.

Node 2

Hot failover. Full KeyDB replica set. Independent Singularity. Read/write capable.

Node 3

Third replica. PostgreSQL backup. ZFS archive storage. Full data redundancy.

Active-Active, Not Active-Passive

All three nodes accept writes simultaneously. KeyDB handles conflict resolution. This means if Node 1 goes down, Node 2 doesn't just "take over" — it was already serving the whole time. Zero switchover latency.

How a Session Works

Here's what happens when an AI session starts with Lightning Memory:

1. Load Conscience

The AI loads its project's conscience from Redis. This contains operational rules, current state, infrastructure reference, and session context. Instant.

2. Recall Memories

Project-specific memories are searched. Recent work, decisions, preferences, and context flow into the session automatically.

3. Cross-Project Search

If needed, Unimatrix provides semantic search across ALL projects. "What did we decide about failover?" finds answers wherever they live.

4. Session Archival

Continuum streams the conversation to persistent storage in real time. New memories are saved. Conscience is updated. Nothing is lost.

Memory Database Map

Each project and person has their own isolated Redis instances. Here's the current allocation:

ScopeMemory DBConscience DBPurpose
User 1 (Internal)DedicatedDedicatedPersonal context, preferences, decisions
User 2 (Internal)DedicatedDedicatedPersonal context, preferences, decisions
User 1 — Client 1DedicatedSharedClient platform context
User 2 — Client 1DedicatedSharedClient platform context
Project: ITSMDedicatedDedicatedService management development
Project: PlatformDedicatedDedicatedCore platform development
Project: ValidationDedicatedDedicatedCode validation engine
User 2 — Customer 1IsolatedIsolatedCustomer-specific project context

Related Components

Lightning MCP

The platform that powers memory. 180+ tools through a single MCP endpoint.

Lightning Continuum

Streams conversations to memory in real time. Never lose a session.

Daily Planet

AI reporters that use Unimatrix to search across all projects for monitoring.

Lightning ITSM

Knowledge base and archive tier. 500+ articles, full audit trail.

Ready for AI That Remembers?

Lightning Memory is available as part of the Lightning MCP Platform. No vendor lock-in, runs on commodity hardware.

Contact Us Learn About Lightning MCP