Sovereignty, provenance, and a human in the loop — by default, not by request.
Everything on this page describes how Eclypse actually runs, in every deployment we operate — including fully air-gapped, with no outbound connectivity and no public-model training. It is not a tier you upgrade into — it is the architecture.
Deployed inside your perimeter, not alongside it.
The engine, the Validator Framework, and the registry install as one system inside whichever boundary your data already lives behind.
Sovereign cloud
In-country tenancy under your account and controls, including Singapore's Government Commercial Cloud.
On-premise
Runs on your hardware behind your firewall, integrated with your identity provider and logging stack.
Air-gapped
No outbound connectivity. Local model serving, with module updates applied on your schedule.
Site-resident data
For multi-site research, source data stays with the originating site; only de-identified, permitted derivatives move.
Role-based, by module
Every module declares the data it may touch. Agents and users can only call what their role permits — permissioning is a property of the registry, not a policy layered on top.
Immutable, per run
Every agent run is logged: which modules fired, what evidence was retrieved, which checks passed, and who signed off. The log is not editable after the fact.
Versioned, not silent
A module change ships as a controlled release with a diff, never a silent prompt edit. Outputs stay reproducible against a known version.
Your data never trains a public model. Full stop.
This is the one commitment every deployment topology shares, regardless of how "closed" the environment already is. Nothing your team or Eclypse runs through the platform is used to train a public or foundation model — not for fine-tuning, not for evaluation, not by default and not by exception.
Explainable by construction, not by explanation after the fact.
Human-in-the-loop, structurally
Approval gates are part of the workflow definition, not a setting someone can switch off under deadline pressure. Nothing is auto-selected or auto-published — a named specialist reviews reasoning and sources before a deliverable is used.
Dual-model verification
A second model class, plus rule-based checks, tests claims, statistics, and references before a draft ever reaches the expert workspace. Verification is a pipeline stage, not a hope.
Provenance on every fact
The Validator Framework tracks where each fact came from and when it changed. Every output resolves to its source document, version, and the agent run that produced it — traceability is a property of the data model, not a report generated afterward.
Scoped, registered agents
Agents are constrained to their registered tools and declared data scope. An agent cannot reach past what its role permits, and every capability it calls is versioned and reviewable.
Designed from the ground up for the most demanding regulatory and data-residency requirements.
Common questions about security and governance.
Does Eclypse train public or foundation models on our data?
No. Client operations and proprietary data are never used to train a public or foundation model, in any deployment topology — sovereign cloud, on-premise, air-gapped, or site-resident.
Where does our data actually live?
Wherever you choose: in-country sovereign tenancy (including Singapore's Government Commercial Cloud), on your own hardware behind your firewall, fully air-gapped with no outbound connectivity, or resident at the originating site for multi-site research. The default is that data does not leave your environment.
Is a human always in the loop?
Yes, by design. Every deliverable is reviewed and signed off by your specialist before it is used; nothing is auto-selected or auto-published. Agent outputs pass a second model class plus rule-based checks before a human ever sees them.
Can we trace an output back to its source?
Yes. Every output resolves to its source document, version, and the agent run that produced it, through the Validator Framework's provenance tracking and an immutable audit log.
What compliance standards does Eclypse align with?
ISO 27001 and SOC 2 alignment, GDPR and PDPA compliance, and HIPAA / BAA readiness, with role-based access control by default.
Read the full platform architecture, or bring us your constraint.
ECLYPSE
AI
Our proprietary AI orchestration platform for healthcare: one engine, a registry of reusable task modules and domain agents, and a governed knowledge base — deployed inside your walls and run by your team.