A structured knowledge graph of everything APMEX has built in the Knowledge Center, delivered as a perpetual asset. Numis is named after numismatics - the study and collection of coins and currency - a nod to APMEX's core identity as the world's largest online precious metals retailer.
8,000+
KC content assets mapped
5,000+
CoinGrader+ entries
6+
Tools and calculators
Perpetual
License - no subscription
What is Numis
Numis is a knowledge graph built from the content APMEX already owns. Every article, guide, video transcript, tool, calculator, and CoinGrader+ entry in the Knowledge Center is processed and stored as a structured graph node - with explicit relationships connecting articles to topics, topics to metal types, content to tools, and pages to each other.
Unlike a traditional CMS or search index, a knowledge graph captures meaning and context. It knows that a beginner silver article is related to three calculators, is missing an internal link to a coin grading guide, and is part of a topic cluster that is underserved relative to the site's gold coverage. Those relationships are queryable and actionable - by your team, by AI systems, and by other applications across APMEX's infrastructure.
APMEX receives the graph as a perpetual licensed asset. It is portable, not tied to any subscription, and built to be extended as the KC continues to grow. Numis is not a replacement for your CMS - it is the intelligence layer that runs alongside it.
What's Inside the Graph
Full Content Graph
All 8,000+ assets as graph nodes with explicit relationships: article to topic, topic to metal type, content to calculator, author attribution, and internal link map. Every piece of content has a defined place in the knowledge structure.
SEO Signal Layer
Ranking topics, traffic benchmarks, and content gap analysis baked into the graph. Know which topics are well-covered, which are underserved, and where a single new article would have the highest impact on organic visibility.
Queryable at Launch
Ask questions the current CMS cannot answer: "Which silver articles have no internal links to the coin calculator?" or "Which beginner topics have fewer than 3 articles?" The graph answers in seconds. No custom SQL. No report request to a dev.
Perpetual License
APMEX owns the knowledge graph snapshot outright. It never expires and is not tied to a subscription. The graph is delivered as a portable asset alongside your platform rebuild, and remains yours regardless of future vendor relationships.
Agentic Foundation
The knowledge graph is designed as the source of truth for AI-powered workflows. Once built, it powers agents that answer customer questions, surface related content automatically, generate editorial briefs for gaps, or flag when high-traffic articles go stale. This is the infrastructure layer that makes AI useful inside the KC, not just bolted on top.
Integration Ready
Built with an API layer so it can be queried by other APMEX systems. Product pages can surface related KC content. Customer-facing search can rank results using topic relationships. Internal teams can pull content intelligence into reporting workflows. It is a connected asset, not an isolated database.
What Becomes Possible
AI-powered customer recommendations
Surface related KC content to customers browsing product pages - matched by metal type, skill level, and purchase intent. No manual curation required.
Predictive editorial planning
Identify topic clusters where content is thin relative to search demand. Prioritize what to write next based on coverage gaps and traffic signals already in the graph.
Internal operations intelligence
Editors can query the graph to find duplicate coverage, articles missing schema markup, or content that references outdated pricing. Operations workflows become data-driven.
Cross-application data layer
Connect Numis to product pages, CRM, and search. When a customer views a Gold Eagle coin, the graph knows exactly which KC articles, guides, and calculators are most relevant to their session.
Corporate-wide AI source of truth
A single structured knowledge base that any AI system inside APMEX can query. Eliminates the risk of different teams training AI tools on conflicting or outdated content sources.
Customer-facing AI tools
Power chat assistants, smart search, and guided buying tools that answer customer questions using verified KC data - not hallucinated responses from a general-purpose AI with no domain grounding.
Brain Architecture
How APMEX connects to the intelligence layer
Numis is the entry point. Once the Knowledge Center is mapped as a graph, every data source and company function that touches content becomes a potential node. The diagram below shows the full picture: what feeds in, what the intelligence layers do with it, and what surfaces benefit on the other side.
Numis Knowledge Graph (core)
Data inputs
Brain intelligence layers
Connected surfaces
Data That Feeds In
Knowledge Center
8,000+ articles, videos, calculators, and topic relationships. The primary corpus. Every asset is a node; every internal link is an edge.
Product Catalog
Coins, bars, rounds, and collectibles. Products linked to relevant KC education so customers see context alongside the buy button.
CoinGrade+
5,000+ graded coin records with condition, valuation, and type data. Extends the graph's numismatic depth beyond editorial content.
Live Spot Prices
Gold, silver, platinum, and palladium feed. Price signals mapped to investment articles and calculators so content responds to market conditions.
Search Query Log
What customers search for vs. what the KC covers. The gap between search intent and content coverage is the most actionable editorial signal in the graph.
Behavioral Signals
Video plays, calculator usage, article dwell time. Engagement signals mapped onto graph nodes to surface what is resonating vs. what is ranking without being read.
Company Areas That Benefit
E-Commerce
Product Pages
Product pages surface related KC education automatically based on graph relationships. A customer looking at a Morgan Silver Dollar sees the relevant grading guide, not a generic "learn more" link.
Content
Editorial Workflow
Content team workflow powered by gap analysis and the SEO signal layer. Editors know exactly which topics to prioritize before a competitor fills them.
Support
Customer Service
AI-assisted responses grounded in verified KC content. No hallucination risk because answers trace to specific graph nodes your team has already published and approved.
Legal
Compliance
Every article referencing a specific product claim, storage guarantee, or return policy is queryable in seconds. Audit prep that used to take days takes minutes.
Growth
Marketing
SEO gap analysis and content performance signals drive campaign prioritization. The graph knows which topics are winning organically and which are leaving traffic on the table.
Revenue
Sales & CRM
Customer content behavior wired to sales outreach. A prospect who has read three IRA rollover articles in a week is a different conversation than one who hasn't. The graph surfaces that signal.
Brain Infrastructure
Seven layers beneath the graph
Numis sits at L2. The six layers around it are what turn a knowledge graph into a compounding intelligence platform. Each layer has a distinct job. Together they are what separates a queryable graph from an AI-ready infrastructure stack.
L7
Learning Membrane
data sovereignty · cross-tenant boundary · anonymized patterns only
The governed boundary between APMEX's tenant namespace and the broader brain platform. Only anonymized patterns cross this line: no raw content, no customer data, no PII. Every cross-tenant promotion requires human approval. APMEX's intellectual property stays inside the boundary.
L6
Training Pipeline
domain classifier · multi-turn windowing · Noema model training
APMEX's interactions with the graph feed a domain-specific training pipeline. Editorial decisions, content gap resolutions, and AI query outcomes are logged, classified, and used to fine-tune models scoped to the precious metals domain. The platform gets smarter about APMEX's subject matter with every decision made on top of it.
Consent records, deletion attestations, and retention policies enforced at write time. Relevant for precious metals content given FTC disclosure requirements and CFTC-adjacent investment guidance. If APMEX ever puts a customer-facing AI in front of users, this layer handles the legal exposure before a single query is answered.
The licensed feature set available to APMEX as a tenant. Which intelligence capabilities are active, which are reserved for future phases, and which are available as add-ons. Entitlements are managed centrally; new capabilities are activated without touching the underlying graph or frontend.
Per-user memory scoped to APMEX's tenant namespace. A returning customer who has read three silver IRA articles picks up where they left off. A customer service rep who opened a compliance query last week sees its context resurface automatically. Memory is what makes the AI layer feel like a system that knows you, not a search box that resets every visit.
L2
Knowledge Graph: Numis
8,000+ content nodes · topic relationships · SEO signal layer · content gap map
The entity-relationship core. Every KC article, video, calculator, and topic is a node. Every relationship between them is an edge. SEO signals, traffic patterns, and content gap data are properties on those nodes. This is what makes recommendations explainable: not "the model says so" but a traversable path through APMEX's own content history and topic structure. This is Numis.
L1
Inference Router
AI Router · model selection · entitlement enforcement · usage metering · failover
Every AI query routes through the model router. Model selection, entitlement checks, usage metering, and multi-model failover all happen here before a single token is generated. APMEX's tenant tier determines which models are available and at what cost. Routing decisions are logged and every query is auditable.
Intelligence Trajectory
What the graph looks like in year three
Numis on day one knows everything the KC has published. That is already more structured intelligence than any competitor has about their own content. What happens after that is where the real value builds.
Day One
The baseline
8,000+ content assets mapped. Topic relationships established. SEO signal layer active. Content gap map generated. Editors can ask questions the CMS has never been able to answer.
Year One
The editorial layer
Gap analysis has driven content decisions. New articles have been created to fill high-traffic topic voids. The graph knows which editorial decisions produced organic traffic and which filled gaps without moving the needle. Customer AI queries have started logging which graph nodes answer questions well.
Year Three
The compounding asset
The graph reflects three years of editorial decisions, customer behavior, product catalog evolution, and market movements. Fine-tuned models have been trained on APMEX's precious metals domain. The gap between what APMEX's AI can answer and what a generic model can answer is structural and widening.
JM Bullion and Kitco have content. They do not have three years of mapped topic relationships, SEO signal data, and customer query history feeding a domain-specific intelligence layer. That gap does not close by adding more articles. It closes by building this infrastructure, and the sooner it starts compounding, the wider the moat.
What APMEX Owns
Sovereignty by design
The brain infrastructure is built so APMEX owns what matters. The data, the graph, and the intelligence layer are assets, not a subscription to someone else's platform.
Graph Sovereignty
The Numis knowledge graph is APMEX's asset. On the license path it is hosted on Swift Innovation infrastructure, but the data, relationships, and SEO signals belong to APMEX. On the build path it is self-hosted with no platform dependency at all. Either way, the graph does not disappear if the vendor relationship ends.
Perpetual License
The graph is delivered as a permanent asset under a perpetual license. No annual renewal, no per-query fee for the graph itself. APMEX pays once to map the KC into the intelligence layer and owns that mapped state indefinitely. Future model improvements and expanded brain layers are available as optional additions, not required upgrades.
Content Corpus Portability
The structured content corpus (articles, topics, relationships, and SEO signals) is exportable in standard formats. If APMEX ever migrates platforms, changes vendors, or builds an internal AI team, the corpus travels with them. The intelligence layer is not locked to any CMS, any model provider, or any vendor relationship.
The brain licensing model means APMEX gets infrastructure we have already spent two years building and testing, delivered faster and cheaper than a ground-up build. What APMEX builds on top of it (the editorial decisions, the customer queries, the domain-specific training data) is exclusively theirs.
Delivery by Path
Custom Rebuild
Built-in to Phase 3
When
Runs in parallel with Phase 3 (headless CMS and content migration). No timeline extension.
How
The graph is built natively into the headless CMS architecture. Content relationships are first-class data, not an afterthought.
API access
GraphQL + REST endpoints delivered alongside the platform. Query Numis from any internal system at launch.
Optional
Add-on to the Custom Rebuild engagement. Provisioned as a licensed brain instance alongside Phase 3; no separate contract required.
WordPress Rebuild
Separate Engagement
When
Scoped and quoted separately after the WordPress platform is live and stable. Typically starts in Phase 3 or after launch.
How
A plugin bridge layer connects the knowledge graph to the WordPress environment. Content relationships are maintained outside the WP database for portability.
API access
Same GraphQL + REST endpoints as the custom path. The graph is platform-agnostic and can be queried from any system regardless of CMS choice.
Standalone
Independent engagement from the WP Rebuild. Can be initiated after launch or deferred to a future phase without impacting the core rebuild timeline.
Investment
Numis Add-on
Brain licensing - Perpetual - Your tenant, your data
Pricing model
Brain licensing fee - not a time-and-materials build. Numis is provisioned as a dedicated brain instance scoped to your domain. Rather than building the graph infrastructure from scratch, APMEX receives a licensed deployment of existing brain infrastructure, adapted to the Knowledge Center and delivered as a permanent asset.
Custom Rebuild
Add-on within the engagement - runs in parallel with Phase 3, factored into the monthly retainer. No separate contract or additional billing period required. Licensing fee confirmed at kickoff.
WordPress Path
Separate engagement after platform launch. Licensing fee and tenant tier confirmed at that time based on KC volume and integration scope.
License
Perpetual, one-time - no annual renewal, no per-seat fee, no usage metering. APMEX owns the graph instance outright at delivery.
Maintenance
The graph is a snapshot at delivery. Ongoing enrichment is available as a monthly retainer add-on or can be handled by APMEX's internal team using documented tooling.
Numis is provisioned as a dedicated brain instance scoped to your domain. There is no ongoing subscription for the graph itself. The licensing fee and tenant tier are confirmed at kickoff.