Canonical Graph
A canonical graph is the authoritative graph of an enterprise's entities and their relationships — one verified identity per real-world thing, with every system, application, and AI agent resolving against it as the single source of truth.
What is a canonical graph?
A canonical graph answers the most fundamental question in enterprise data: "which real-world thing are we talking about?" In every sizable company, the same customer exists five times — as "Acme Inc." in the CRM, "ACME, LLC" in billing, a dormant duplicate in the marketing platform, and two spellings in support tickets. Humans paper over this with judgment. Software cannot, and AI agents least of all.
The canonical graph resolves those duplicates into one node per entity — with a canonical identifier, typed attributes, and precedence rules saying which source system owns which fact — and then goes further: it masters the relationships. Which customer holds which contracts. Which supplier feeds which product line. Which employee owns which account. Those edges are what turns a list of clean records into a map of the business an AI system can actually navigate.
Why does "canonical" matter in the age of agents?
Because agents act on references, and unresolved references are how small data errors become large operational ones. Ask an agent to "email our biggest churn-risk customers," and its very first step is resolving which customers those are. Resolve wrong, and everything downstream — the audience, the offer, the tone — is wrong with total confidence. Unlike a bad report, which a human squints at and discounts, a bad agent action executes.
This reframes entity resolution from a data-quality chore into safety infrastructure. Enterprises building what vendors now call an AI execution plane consistently arrive at the same foundation: before agents can reason over the business, the business must exist as canonical, deduplicated, relationship-typed entities. The graph is the form that foundation takes.
How does a canonical graph compare to a warehouse, MDM, or catalog?
| System | Core job | Where it falls short for AI |
|---|---|---|
| Canonical graph | One verified identity per entity, plus mastered relationships | — (built for exactly this) |
| Data warehouse | Historical analytics over copied data | Duplicates survive ingestion; no live entity identity |
| MDM platform | Golden records for individual entities | Relationships are second-class; batch-oriented |
| Data catalog | Inventory of datasets and their metadata | Describes data about things; never resolves the things |
None of these disappear. The warehouse still analyzes, the catalog still inventories. The canonical graph is the layer they have all been missing: the resolution authority that says the row in the warehouse, the record in the CRM, and the ticket in the support queue all refer to this customer.
What canonicalgraph.com is building
canonicalgraph.com is an early-stage property in build by ROIZILLA, LLC. The planned direction is reference architecture and tooling for canonical entity resolution — identity schemes, precedence models, and resolution workflows that agents can query. That capability is not live today. This page is an educational reference on the concept; no products, services, or customer features are offered yet.
Frequently asked questions
Is a canonical graph the same as MDM?
It is the graph-native successor. MDM masters individual records; a canonical graph also masters the relationships between them — the information agents actually navigate.
Does it replace systems of record?
No. CRM, ERP, and billing remain where work happens. The canonical graph sits above them as the resolution layer, assigning canonical identities and attribute-level authority.
Why do AI agents need one?
Agents act on entity references. Resolving "our biggest customer" to the wrong record corrupts every downstream action. The graph answers "which real-world thing is this?" before the agent acts.
How is a canonical graph built?
By continuous entity resolution: matching and merging duplicates across systems, assigning canonical IDs, typing relationships, and re-resolving as sources change. The discipline matters more than the database.
What happens when two systems disagree about a fact?
Precedence rules in the graph decide which system owns which attribute; unresolved conflicts surface as reviewable events instead of silent inconsistencies.
Who operates canonicalgraph.com?
canonicalgraph.com is owned and operated by ROIZILLA, LLC. It is a property in build, and this page currently serves as an educational reference.
Related properties in the ROIZILLA network
- enterprisegraphs.com — enterprise knowledge graphs as execution infrastructure
- contextspec.com — the specification layer for enterprise context
- decisiongraphprotocol.com — the protocol for governed AI decision graphs
- graphcontextlayer.com — graph-structured context for AI systems