Your content and data models, connected. So AI can build across them.
Look up before you make up.
ARAMAI is a semantic intelligence company. CoreModels is where your agents build and operate structured data — import it, map it, validate it, generate it back — from one governed model your team edits. Every write is versioned and reviewable. Every answer resolves to the model.
Trailer transcript. Look up before you make up.
- > what was completed revenue last month?
- $ dbt parse
- > which status list drifted?
- > who owns what status means?
- > map customer across four systems
- > what was completed revenue last month?
AI builds. Humans govern. CoreModels connects both.
What your agents do with CoreModels
Agents build.
An agent creates and evolves types, mappings and validation against the governed model, then generates the artifacts your stack needs.
Agents operate.
An agent imports what dbt already writes, audits drift in CI, answers what breaks, and writes the contract back.
Agents answer.
An agent answers what a field means, which values are legal and who owns it — from the model, not from inference. Measured by the unattended rate: the share of your agents' answers you can act on without a person checking them.
What becomes possible
Your systems already hold years of settled meaning: what an account is, what an order means, which content reaches which audience. Connect those definitions, and every AI agent you run can build across all of them at once.
YOUR SYSTEMS, AS THEY ARE
account
CRMthe company that receives the invoice
customer_id
warehousethe person an order ships to
CORE MODEL
customer
two definitions, one mapping, a note on when to use which
one named owner
WHAT YOUR AGENTS BUILD
- a revenue report
- a support answer
- a new integration
Your systems stay the authority for what they hold. CoreModels holds the mappings between them, and the decision when they disagree.
The platform: CoreModels
CoreModels holds the mappings and governed meaning between your data and content platforms, warehouse, API and ontology schemas — and the public standards they connect to. Whatever you build or transform with AI works from the connected picture: no manual stitching, no guessed structure, no hallucinated mappings. Pipelines, integrations, agents and apps all work from one Core Model.
1
Connect your models.
Import a dbt manifest, a warehouse catalogue, a JSON Schema, a Schema.org vocabulary or a SKOS taxonomy. Nothing to migrate.
2
Map it together, with AI.
AI proposes the joins. A person on your team accepts, edits or rejects each one, and every change is recorded.
3
Point your agent at it.
Claude, Cursor or your own agents read the model through the CoreModels MCP server.
Your source systems stay as they are. Your team decides what terms mean: AI proposes, a person commits. Your agents' prompts and traffic stay yours and are not used to train models.
What changes when AI can look things up
92%
Accuracy with schema-aware retrieval in production deployments, against 68% for vector-only retrieval
90%
Reduction in AI hallucination on tested workloads
5–16×
Token efficiency on the CoreModels MCP benchmark
In the open: Sage Bionetworks publishes its Synapse data-model registry into CoreModels from a public GitHub pipeline. See the repository →
Two decades refining the methodology.
The CoreModels approach grew out of twenty years of schema virtualization and mapping work across healthcare, finance, content, and enterprise systems.
The organizations below have implemented the methodology in production.
A demo built around your systems
Answer a few questions: what AI should build, and which systems it should build across. We'll assemble a free video tour of how your entire AI program gets a power-up from the structured content and data assets you already have, and email it to you.
A person on our team records a 3 to 5 minute video on the systems you name and emails it within two business days. No sales call unless you ask for one.