A short reference on how Pacts work.
A Pact is an optional federation layer for credit unions running CUcomputer. This page explains what a Pact is, what is exchanged across one, and how the economics differ from building AI infrastructure alone. There are no calls-to-action on this page.
What a Pact is.
A Pact is a federation of three to ten non-competing credit unions that have chosen to share learnings derived from their AI Agents. Each credit union runs its own CUcomputer deployment in its own environment; the Pact is a shared learning layer on top, not a shared system.
Pacts are opt-in. CUcomputer works without one. A credit union can deploy CUcomputer privately, generate its own intelligence in isolation, and never join a Pact. Members of a Pact can leave at any time without affecting their underlying deployment.
Eligibility is geographic and segment-based: a Pact is composed of credit unions that do not compete for the same members. Two CUs in the same metro typically cannot be in the same Pact; two CUs in different states can.
What gets shared.
Patterns travel. Member-identifiable data does not.
Shared across the Pact
- +Detected fraud signatures (vector representations, no PII)
- +Decision-logic patterns that produced verified outcomes
- +Marketing playbooks that were approved and converted
- +Anti-patterns — approaches that failed, with cause attribution
- +Aggregate benchmarks at the cohort level
Never shared
- −Member names, addresses, dates of birth, or any direct identifier
- −Account numbers, balances, or transaction-level records
- −Documents, statements, agreements, or scanned files
- −Free-text notes, support tickets, or correspondence
- −Model weights or training data of the originating CU
Pact exchange is implemented as cryptographically-bounded learning units. The originating credit union keeps the raw data; only the abstracted pattern leaves the environment, and only after verification. This is consistent with the federated-learning architecture defined in NIST AI 600-1 §3.2 and aligned with NCUA guidance on third-party data exchange.
The cost of going alone.
For context, not for sale.
Solo cost ranges are drawn from publicly reported credit union AI initiatives in the US and Canada (2023 – 2026). Engineering maintenance assumes a two-to-three person team. Pact members do not pay an additional Pact fee on top of CUcomputer; the federation is structural, not commercial.
The premise of a Pact is mathematical, not philosophical: the rate at which a model improves is a function of the diversity and volume of verified signals it observes. A single credit union sees its own signals. A federation of ten sees ten times the diversity, with attribution preserved. Quality compounds; cost does not.
Questions about Pacts or the federated-learning architecture can be sent to the team at cucomputer.com/contact. This page is informational; it is not part of any sales process.