AVIA · Ontology
A digital twin of your worldyour agents can act on
AVIA Ontology turns your data into Objects, Links, and Actions — typed digital twins of real entities, the relationships between them, and the write-back operations that change the real world. Both a semantic layer that describes your world and an operational layer your agents act on.
The model
Objects Links and Actionsnot a passive graph
Like Palantir's ontology, AVIA models your business as Objects (customers, products, orders, devices) carrying Properties, joined by typed Links, and operated on by Actions. Agents and people don't just read it — they change the world through it.
Objects
Digital twins of real entities — every customer, order, product, and device as a typed object.
Properties
The facts each object carries — status, quantity, location — kept current and governed.
Links
Typed, directional relationships — an order places a product, a shipment ships from a site.
Actions
Write-back functions that change the real world — and the object's properties with it.
What's different
Agents build and manage the ontologythen act on it
Most knowledge graphs are hand-modeled once and left to rot. AVIA's agents resolve entities, propose and curate typed links, and keep the graph governed and current as the world changes — then query it and take Actions on it. The ontology stays alive because the same agents that read it also maintain it.
- Agents resolve and de-duplicate entities into objects
- Agents propose, curate, and repair typed links
- Agents keep properties current and the graph governed
- The same agents query the graph and take Actions on it
Built multimodally, reused everywhere
One product ontologyone user ontology
Built multimodally from AVIA's annotated data, the ontology unifies a product layer and a user layer that are reused across products — the shared knowledge behind ProductFlow's product intelligence and HealthDoor's personal health memory.
Product ontology
Products, variants, attributes, and their relationships — the structured product knowledge that powers ProductFlow.
User ontology
People, their context, and history as typed objects — the personal health memory behind HealthDoor.
How it works
Inputs become a governedoperational knowledge layer
Images, documents, and logs are resolved into objects and typed links, organized into a governed graph, and served by API, RAG, query, and export — with Actions wired back to your systems.
Resolve objects
Agents identify and de-duplicate entities into typed objects across every source and modality.
Model links
Capture the typed relationships that connect them into a coherent, queryable graph.
Govern
Versioning, access control, and lineage keep the graph trustworthy and auditable.
Serve & act
Expose it by API, RAG, query, and export — and run Actions that write back to your stack.
Capabilities
A knowledge layer your AI can rely onand act through
Object modeling
Resolve entities across modalities into typed objects carrying governed properties.
Typed links
Explicit, directional relationships agents can traverse and reason over.
Actions & functions
Write-back operations that change the real world — and the objects with it.
Agent-driven curation
Agents resolve objects, propose links, and keep the graph current as the world changes.
Governance & lineage
Versioning, access control, and full lineage make every fact auditable.
Multimodal & served
Built from images, documents, logs, and databases; served by API, RAG, query, and export.
In practice
Where it earns its keep
Grounded enterprise RAG
Retrieval grounded in typed objects and links, not loose text fragments.
Agents that take action
Agents query the graph, then fire Actions that update orders, devices, and records.
Analytics & search
Expose objects and relationships to analytics, search, and operational dashboards.
By the numbers
Knowledge agents can calland act on
1
Source of truth
Act
Write back to the world
API
Query the graph
BYO
Storage & keys
Built to integrate
Runs in your environment
BYOS
The graph and its sources stay in your storage.
BYOM
Use your own models for resolution and embedding.
BYOK
Your keys, access control, and full lineage.
FAQ
Ontologyanswered
A vector store finds text that looks similar; it has no concept of an entity. The ontology models typed objects, explicit relationships, and properties — and adds Actions that write back to the real world. Agents reason over structure and operate on it, not just retrieve nearby chunks.
Give your agentsa world they can act on
AVIA Ontology turns scattered data into Objects, Links, and Actions — a governed layer your AI can query and operate.