Custody Reference Architecture
Wallet tiers, HSM/MPC boundaries, policy engine, approvals and recovery.
Explore solution →Architecture frameworks that expose trust boundaries, service separation, integrations and operating models instead of relying on sales claims.
Each architecture is adapted during discovery to the institution, regulation and existing systems.
Every design is evaluated with trust boundaries, failure modes, operating responsibility and handover model.
Bring the current system, integrations and target operating model to an engineering session.
The messaging layer can use enterprise event-bus technologies such as Kafka/RabbitMQ, while signing can be designed around HSM, MPC or hybrid models based on the client’s security and procurement requirements. This is a reference architecture and does not claim use of a particular vendor or protocol.
These are technology families that can be evaluated during solution scoping; this is not a claim that every item is pre-enabled in every project. Exact network, token-standard and integration scope is defined during technical discovery.
Protocol families such as GG18/GG20/CMP and quorum models such as 2-of-3 or 3-of-5 are evaluated against the threat model, vendor selection, recovery design and client key-ownership requirements. Selection is documented through architecture decisions and acceptance criteria rather than implied as a universal implementation.
Crypto Software delivers client-specific products on reusable engineering components and reference architectures. Source-code/IP transfer, on-prem/private-cloud deployment or managed operations are defined explicitly per engagement.
Client-specific build with source, documentation, IaC and operating runbooks handed over after acceptance.
Deployment in the client data centre or private-cloud account; keys, data and infrastructure can remain client-owned.
Managed operations with explicitly scoped SLA, observability, incident/recovery and change management.
Source-code delivery does not mean rebuilding every system from scratch. Reusable policy, API, auditability, integration and operating components are adapted to client security, residency and process requirements. “Independently audited” is stated only where a verifiable report exists for the relevant version and scope.
This is a reference mockup illustrating the expected enterprise control-plane UX, not customer data or a claim of a live product screen.
Actual SLA, RTO and RPO values are committed only after deployment model, region count, database/replication design and client risk/budget requirements are defined. Unverified percentages or minute values are not published.
Each step is designed to be traceable with correlation IDs, idempotency and audit events. Actual topology varies by client trust boundaries and residency requirements.
Example: a high-value or new-destination transaction can require extra approval, allowlist validation, risk checks or a time lock. Threshold values belong to client policy.