Your AI agents don't know who the customer is.
hubgentic is the entity layer for enterprise agents. It resolves the same customer across core banking, CRM, KYC and support systems at query time — and tells your agent exactly where every field came from. No data lake. No copies at rest.
Free sandbox with realistic mock systems · No credit card · MCP + REST
UnifiedEntity Counterparty
├─ legal_name
│ ├─ ★ 'BIANCHI LUCIA' ← core_banking.CIF[CIF-1003].NAME_TXT rule=fuzzy.name_dob_email conf=1.00
│ └─ 'Lucia Bianchi' ← crm.Account[001A000003].Name rule=fuzzy.name_dob_email conf=1.00
├─ national_tax_code
│ └─ ★ 'BNCLCU85M41F205X' ← core_banking.CIF[CIF-1003].TAX_CODE # CRM never had it
├─ date_of_birth
│ ├─ ★ '1985-08-01' ← core_banking.CIF[CIF-1003].BIRTH_DT
│ └─ '1985-08-01' ← crm.Account[001A000003].Birth_Date__c
└─ email
├─ ★ 'lucia.bianchi@gmail.com' ← core_banking.CIF[CIF-1003].EMAIL_ADDR
└─ 'Lucia.Bianchi@GMAIL.com' ← crm.Account[001A000003].Email__cAgents are only as good as the context you can prove.
Customer-facing agents answer from whichever system they were wired to. In a bank, an insurer or a telco, the truth is split across a dozen systems that disagree — and you can't just copy it all into one place.
Same customer, five records
"ROSSI MARIO" in core banking, "Mario Rossi" in the CRM, a typo in the ticketing tool. Your agent sees three strangers.
No lineage, no trust
When an agent states a customer's tax status, compliance asks: which system, which field, fetched when? Today nobody can answer.
Data can't move
Banking secrecy, residency and supervisory rules make "just centralise it" a two-year program. Agents need answers now.
How it works
Define the business entities once. hubgentic binds them to your systems of record and does the hard part on every request.
Model
Describe entities like Counterparty in a vendor-neutral ontology — attributes, identifiers, relationships.
Bind
Map each attribute to native fields in every source: CIF.NAME_TXT, Account.Name. No source is privileged.
Resolve
At query time, fetch from all sources concurrently and correlate records with deterministic, normalized and fuzzy rules.
Explain
Return one unified entity where every value carries its source, field, timestamp, matching rule and confidence.
Built for agents. Built for auditors.
Federated entity resolution
Clean shared keys, dirty identifiers, no shared key at all — each match states the exact rule that fired and why.
Field-level lineage
Provenance is part of the data contract. It is impossible to emit an unattributed value.
MCP-native
Plug into Claude, your own agent framework, or any MCP client. REST API with workspace-scoped keys too.
Governance per workspace
Every source fetch passes through a policy check. Switch systems off per workspace; denials are explained in the response.
Nothing at rest
Correlation is request-scoped. Explanations live in a short-lived cache, then disappear.
Entity-anchored memory Exploring
Agent memory keyed to the resolved customer, not the chat session — so every agent remembers the same person. Tell us if you need this.
Help us build the layer your agents are missing.
We're looking for a handful of teams running agents against messy systems of record. Try the sandbox, then tell us what breaks.