Architecture
Ontic sits between your app and your model. Every claim is checked against your data. Evidence found: authorized. Evidence missing: blocked. Everything logged and signed.
Need vocabulary first? See Gate and Fails closed.
How the gate runs
Read left to right: request enters, authority is checked against registered evidence, and only authorized output leaves the system.
The same flow applies in every environment; deployment tier changes where enforcement and attestation occur.
This boundary prevents category drift: Ontic is an enforcement layer, not a model or post-hoc monitor.
Ontic is
- • A gate that checks claims against real data
- • Deterministic — same input, same decision, every time
- • An audit trail you can hand to anyone
Ontic is not
- • A model (we don't generate anything)
- • A content filter (we don't judge tone or topic)
- • A monitoring tool (we prevent, not just detect)
- • A replacement for your data (we check against it)
Product Architecture Matrix
How governance scales from shared SaaS to air-gapped hardware.
Ontic IP (The Governance Engine)
Why do we need you?
Because only we provide the Forensic Chain of Custody, Data Sovereignty, and Mission Logic enforcement for high-stakes environments.
Compare all tiers on desktop.
Public SaaS (Shared)
Target Industry
Prosumer / Creative
Individual Ad-Hoc Use
Ontic Needed?
Risk Profile
Unprotected
Zero Liability Coverage
Data Sovereignty
Ownership
Provider Owned
Input trains their model.
Runtime Isolation
Tenancy
Shared
Multi-tenant, no isolation.
Context Authority
Oracle
Influence
Hope the model listens.
Prompt Engineering
Mission Logic
User Only
System Prompt hidden.
Ontology
Epistemology
Not Available
Safety Policy
The Law
Provider Default
Generic: "No hate."
Guardrails
Policy Enforcement
None
State Extractor
Dispatcher
None
One size fits all.
Tool Sandbox
Action Control
Chat Only
Text generation.
Chain of Custody
Forensics
Black Box
No visibility.
Observability
Ops
None
Human-in-the-Loop
Oversight
None / User Discretion
Caveat Emptor.
Lifecycle Ops
Updates
Continuous (SaaS)
Provider pushes updates.
| Deployment / Product | Public SaaS (Shared) The Studio | Public PaaS (API) The Refinery | Private Cloud (Dedicated) The Clean RoomDedicated | Secure Edge (Air-Gapped) The Clean RoomAir-Gapped |
|---|---|---|---|---|
| Feature | Public SaaS (Shared) | Public PaaS (API) | Private Cloud (Dedicated) | Secure Edge (Air-Gapped) |
| Target Industry | Prosumer / CreativeIndividual Ad-Hoc Use | General EnterpriseMarketing, HR, Support | Regulated IndustryFinance, Legal, Healthcare | Defense / Intel / Critical InfraMilitary, Energy, NatSec |
| Ontic Needed?Risk Profile | UnprotectedZero Liability Coverage | OptionalReputational Risk | CriticalRegulatory / IP Risk | MandatoryLife & Safety / Kinetic Risk |
| Data SovereigntyOwnership | Provider OwnedInput trains their model. | Provider RetentionLogged for 30 days. | Tenant IsolationYour data, their cloud. | Sovereign / On-PremPhysical Custody of Weights/Data. |
| Runtime IsolationTenancy | SharedMulti-tenant, no isolation. | Logical IsolationTenant partitioning. | Hard IsolationDedicated instances, no co-tenancy. | Physical IsolationOwned hardware, air-gapped. |
| Context AuthorityOracle | InfluenceHope the model listens. | Strong ContextInjected via System Prompt. | Total ControlLogits/Decoding constrained. | Total ControlPhysical Data Authority. |
| Prompt EngineeringMission Logic | User OnlySystem Prompt hidden. | User + ConfigurableProvider sees your IP. | Immutable & VersionedProprietary IP. | Immutable & VersionedSigned Firmware. |
| OntologyEpistemology | Not Available | ValidatorChecks output structure. | Arbiter of TruthEnforces logic & consistency. | Arbiter of TruthHard-coded Knowledge Graph. |
| Safety PolicyThe Law | Provider DefaultGeneric: "No hate." | Provider + CustomApp-specific rules. | Custom & AuditedBusiness Logic Enforcement. | Mission CriticalZero-Tolerance Enforcement. |
| GuardrailsPolicy Enforcement | None | Safety NetBlocks known bad inputs. | Deterministic PolicyEnforces business rules. | Deterministic PolicyHardware-enforced constraints. |
| State ExtractorDispatcher | NoneOne size fits all. | Basic ClassificationKeyword routing. | Semantic DispatchIntent-based routing. | Hard-WiredRole-based routing. |
| Tool SandboxAction Control | Chat OnlyText generation. | Function CallingProvider-managed execution. | Sandboxed ExecutionCode runs in isolation. | Read-Only / LocalNo write access / No outbound. |
| Chain of CustodyForensics | Black BoxNo visibility. | Gap at ModelInput/Output visible only. | End-to-EndFull Traceability. | End-to-End + ForensicCryptographic Proof. |
| ObservabilityOps | None | Usage LogsToken counts/Cost. | Deep TelemetryLatency, Drift, Quality. | Mission HealthReal-time Anomaly Detection. |
| Human-in-the-LoopOversight | None / User DiscretionCaveat Emptor. | Async ReviewPost-hoc audit. | Active LearningFeedback tunes model. | The "Kill Switch"Human authorization required. |
| Lifecycle OpsUpdates | Continuous (SaaS)Provider pushes updates. | Version PinnedYou control update timing. | Blue/Green DeploymentTested rollouts. | Secure One-Way TransferPhysical media / Diode. |
The Governance Engine is Ontic’s IP. The table below shows the AI tooling and infrastructure we integrate with at each deployment tier.
See the ecosystem ↓Systems Integration (The Ecosystem)
Is this safe to install?
Yes, because we govern the stack you already run — without rip-and-replace.
Compare all tiers on desktop.
Public SaaS (Shared)
Base Model
Knowledge
Frontier Chatbots
Latest GPT/Claude
Retrieval Stack
Vector DB
Proprietary / Hidden
Provider Browsing
Identity (IAM)
Authentication
Social Login / Email
Google, Microsoft
Policy Engine
Integration
Platform Native
Provider Moderation
Adversarial Defense
Red Teaming
Reactive Reporting
"Report Abuse" button
Infrastructure
Compute
Public Cloud
Shared GPU Fleet
Cost Model
Procurement
Free / Subscription
Per User / Month
Compliance
Standards
ToS
Terms of Service
| Deployment / Product | Public SaaS (Shared) The Studio | Public PaaS (API) The Refinery | Private Cloud (Dedicated) The Clean RoomDedicated | Secure Edge (Air-Gapped) The Clean RoomAir-Gapped |
|---|---|---|---|---|
| Feature | Public SaaS (Shared) | Public PaaS (API) | Private Cloud (Dedicated) | Secure Edge (Air-Gapped) |
| Base ModelKnowledge | Frontier ChatbotsLatest GPT/Claude | Frontier Model APIsCommercial SOTA | Open-Weight Fine-TunesLlama, Mistral class | Distilled / Edge-OptimizedSLMs / Quantized Models |
| Retrieval StackVector DB | Proprietary / HiddenProvider Browsing | Vector DBaaSPinecone, Weaviate Cloud | Enterprise SearchElasticsearch, Milvus | Embedded / LocalLanceDB, Qdrant, FAISS |
| Identity (IAM)Authentication | Social Login / EmailGoogle, Microsoft | API Key ManagementAWS Secrets, Azure Key Vault | Enterprise SSOOkta, Ping Identity | Physical TokenCAC, YubiKey, Biometrics |
| Policy EngineIntegration | Platform NativeProvider Moderation | Orchestration LayersLangSmith, Helicone | Programmable GuardrailsNvidia NeMo | Policy-as-CodeOpen Policy Agent — OPA |
| Adversarial DefenseRed Teaming | Reactive Reporting"Report Abuse" button | Basic FiltersRegex / Keyword lists | Active SimulationGiskard, PyRIT | Hardened EvaluationNIST AI RMF Standards |
| InfrastructureCompute | Public CloudShared GPU Fleet | Virtual Private CloudAWS Bedrock, Azure OpenAI | Isolated InstancesAWS Nitro, Azure Confidential | Owned HardwareNvidia IGX, Dell XR Servers |
| Cost ModelProcurement | Free / SubscriptionPer User / Month | ConsumptionPer Token / Metered | License + ComputeCommitted Spend | CAPEX / Enterprise LicenseHardware + Seat License |
| ComplianceStandards | ToSTerms of Service | SOC2 Type 1Process audit | HIPAA / SOC2 Type 2Operating effectiveness | ATO / FedRAMP HighAuthority to Operate. |
What your auditor gets
Every evaluation produces a signed record: what was asked, what was checked, what was authorized or blocked, and why. The record format is the same across all tiers — what changes is the enforcement locus and the attestation chain.
Start where you are
Teams usually progress in stages rather than jumping directly to the highest-assurance environment.
The Studio (Day 1)
Add labeling and logging to your existing AI pipeline. No model changes. No data source setup.
Effort: Hours
ontic init --template minimalThe Refinery (Week 2+)
Connect a data source. Define required fields. Turn on the evidence check. Try the Oracle Builder to find your authoritative sources.
Effort: Days to weeks
ontic init --template medical # or financial, legalThe Clean Room (When you need it)
Signed everything. Verified execution. For when a court or regulator might ask to see every step.
Effort: Weeks to months
ontic init --template life-safetyOperators and consumers
Each environment changes both the primary operator and the trust expectations of the person receiving the output.
Ready to start?
Know your risk level? Jump into the SDK. Still evaluating? The wizard takes two minutes and generates a report your auditor can read.