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.
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.
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.
Memory is organized into three tiers, each serving a different purpose. Together they ensure nothing is ever lost.
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.
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.
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 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.
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.
Full read/write access. Account manager level. Can create memories, update conscience, manage agents in the target Unimatrix.
Read + scoped write. Project contributor level. Can read all, write only within assigned project scope.
Read only. Observer level. Can search and read customer data for monitoring and reporting. No write access.
Every entity in the system — humans, projects, AI agents — has a unique, human-readable designation that encodes their identity, group, and home collective.
| Group | Count | Members |
|---|---|---|
| Humans | 5 | User A (Project 1, Project 2), User B (Project 1, Project 2, Customer Tenant) |
| Projects | 4 | LMCP, ITSM, DB, Forge — each with own memory + conscience databases |
| Daily Planet | 10 | Clark Mercer, Lois Archer, Chloe Weaver, Jimmy Novak, Perry Ashton, Copy Editor, Archivist, Columnist, Copy Boy, Stringer |
| Infrastructure | 3 | Watchdog (health), Diagnostician (AI diagnosis), Taskmaster (builder) |
| Orchestration | 2 | Arbiter (session broker), Magistrate (task orchestrator) |
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.
Primary node. All 16+ Redis instances. Unimatrix vector DB. Full read/write.
Hot failover. Full KeyDB replica set. Independent Singularity. Read/write capable.
Third replica. PostgreSQL backup. ZFS archive storage. Full data redundancy.
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.
Here's what happens when an AI session starts with Lightning Memory:
The AI loads its project's conscience from Redis. This contains operational rules, current state, infrastructure reference, and session context. Instant.
Project-specific memories are searched. Recent work, decisions, preferences, and context flow into the session automatically.
If needed, Unimatrix provides semantic search across ALL projects. "What did we decide about failover?" finds answers wherever they live.
Continuum streams the conversation to persistent storage in real time. New memories are saved. Conscience is updated. Nothing is lost.
Each project and person has their own isolated Redis instances. Here's the current allocation:
| Scope | Memory DB | Conscience DB | Purpose |
|---|---|---|---|
| User 1 (Internal) | Dedicated | Dedicated | Personal context, preferences, decisions |
| User 2 (Internal) | Dedicated | Dedicated | Personal context, preferences, decisions |
| User 1 — Client 1 | Dedicated | Shared | Client platform context |
| User 2 — Client 1 | Dedicated | Shared | Client platform context |
| Project: ITSM | Dedicated | Dedicated | Service management development |
| Project: Platform | Dedicated | Dedicated | Core platform development |
| Project: Validation | Dedicated | Dedicated | Code validation engine |
| User 2 — Customer 1 | Isolated | Isolated | Customer-specific project context |
The platform that powers memory. 180+ tools through a single MCP endpoint.
Streams conversations to memory in real time. Never lose a session.
AI reporters that use Unimatrix to search across all projects for monitoring.
Knowledge base and archive tier. 500+ articles, full audit trail.
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