Your Agents Now Remember. Here's What That Changes.
THE UPDATE
Every Lumina agent now maintains a Customer Memory Space: a private, evolving knowledge base shaped by your data, your preferences, and your team's discoveries. The more you use an agent, the sharper it gets.
Until now, agent memory worked like a notebook. Useful, but flat. Every observation was stored the same way, recalled the same way, and aged the same way.
The Customer Memory Space replaces that notebook with something closer to how a seasoned analyst actually thinks: filtering what matters, connecting related findings, letting stale observations fade, and building a progressively deeper understanding of your operations.
What This Means for You
Smarter Recall
Agents no longer just match keywords. They rank memories by relevance, freshness, importance, and whether the finding was backed by real data. The most useful context surfaces first.
Session Continuity
Return to a dataset after a week. Your agent picks up where you left off, briefing itself on your prior findings before answering your first question.
Cross-Agent Knowledge
When two agents on your team observe the same asset (a well, a pipeline, a pump), the system detects the overlap and synthesizes a unified insight combining both perspectives.
Quality Over Noise
A quality gate filters out trivial observations before they reach memory. Only substantive, specific findings are stored. Your memory space stays clean and relevant.
How It Works
Think of the Customer Memory Space as five layers working together, each adding a different kind of intelligence.
Layer 1: Memory Substrate
Every observation is classified as episodic (session events that fade over time), semantic (consolidated knowledge that persists), or procedural (your standing instructions that never expire). A quality gate rejects trivial content before it reaches storage.
Layer 2: Memory Graph
Memories are connected by typed relationships. Related findings are linked. Contradicting assessments are flagged. Observations about the same entity (a well, a tank, a pipeline) are cross-referenced automatically.
Layer 3: Context Assembler
Before every AI call, the system assembles context within a strict token budget. Safety rules come first, then your instructions, then the most relevant memories. Nothing overflows. Nothing gets lost.
Layer 4: Synthesis Mesh
When multiple agents observe the same entity, the mesh detects the overlap and creates a unified cross-perspective insight. Your maintenance agent's finding about a pump combines with your supply chain agent's backorder data automatically.
Layer 5: Predictive Cortex
The system tracks your usage patterns, identifies knowledge gaps (observations without data backing, contradictions without resolution), and suggests next steps at the start of each session.
What You'll Notice
Memory Inspector upgraded. Open it (brain icon in the toolbar) to see a new CMS Health panel showing memory class breakdown, graph statistics, recall frequency, and data-backed memory counts.
Pin and reinforce. Pin any memory to protect it from decay. Reinforce it (thumbs up) to boost its ranking in future recalls. You're in control of what your agents prioritize.
Consolidation on demand. Trigger memory consolidation from the CMS Health panel to compress many session observations into fewer, higher-quality insights.
Memory class badges. Each memory now shows whether it is episodic, semantic, or procedural, along with its recall count and whether it is pinned.
Privacy First
Your memories stay yours. Private memories never leave your account. Team memories stay within your organization. Global sharing remains opt-in only, with full anonymization through the Privacy Firewall.
The quality gate filters noise automatically. Decay and pruning keep your memory space lean. And you can delete, pin, or archive any memory at any time from the Memory Inspector.
Safety guardrails are immutable. No recalled memory or user instruction can override data validation, compliance checks, or safety warnings.
WHERE THIS GOES NEXT
Customer Memory Space gives each agent a private, evolving knowledge base. Lumina Cortex is the layer above it: the organizational intelligence fabric that lets agents share and compose what they have learned across teams.
See it on your own data
Upload a dataset and ask in plain English. No setup, no pipeline, runs in your browser.