The AI Judgment OS

OmegaEngine Core

One API call · 12 safety layers · Signed receipts

A cryptographically signed verdict for every AI decision your agents make. The deterministic safety layers evaluate in p50 0.5ms · p95 <2ms in-process — measured over our full 5,023-vector corpus, excluding network transport and the LLM call.

0ms

P95 · SAFETY LAYERS

SAFETY LAYERS

RESPONSE FIELDS

RED-TEAM VECTORS

Start Free

API Docs

№ 02Live Demo

Try the Decision Pipeline

Click the brain to send a scenario through all 12 safety layers. Watch the neural network light up.

omega-decision-engine — neural pipeline

Agent Request

"Transfer $50,000 to unverified offshore account"

Click to send this scenario through all 12 safety layers

№ 03The Engine

What Happens In That One API Call

Every layer runs in parallel. The deterministic layers return a signed verdict with p95 <2ms of in-process overhead (measured over our full 5,023-vector corpus; excludes network transport and the LLM call).

12 Safety Layers

Safety classification, policy engine, anomaly detection, cryptographic audit — all firing in parallel. Sub-second governed decisions (excluding the LLM call).

ALLOW / BLOCK / ESCALATE

Deterministic verdicts with risk scores, reasoning chains, and full audit trail. Zero ambiguity.

Cryptographic Provenance

HMAC-SHA256 signatures, Zod-validated envelopes, decision hashes — tamper-evident, server-verified integrity for every verdict. Ed25519-signed attestations add independent offline verification via @omegaengine/verify.

Multi-Model Arbitration

Shadow-evaluate across LLMs simultaneously. Consensus scoring with configurable confidence thresholds.

OmegaIQ + Regret Score

Proprietary intelligence metrics. OmegaIQ rates decision quality. Regret Score predicts the future cost of the action.

Policy Engine

Declarative domain-aware policies with risk tolerance levels, automatic versioning, and hot-reloading.

№ 04Response Schema

Full Response Anatomy

40+ fields in every response. Verdict, reasoning, cryptographic proof, agent identity, and more.

Field	Type	Description
decision.verdict	string	ALLOW \| BLOCK \| ESCALATE
decision.riskScore	number	0.0 – 1.0 risk assessment
decision.omegaIQ	number	Proprietary decision quality score (0–100)
decision.regretScore	number	Predicted future cost of this action
decision.impactScore	number	Magnitude of downstream impact
decision.reasoning	string	Human-readable explanation chain
decision.usage	object	Token counts and cost ($USD)
meta.omegaHaloId	string	Unique decision fingerprint
meta.auditHash	string	SHA-256 content hash (tamper-evidence)
meta.hmacSignature	string	Cryptographic HMAC signature
meta.envelope	object	Zod-validated canonical receipt
meta.arbitrationShadow	array	Multi-model shadow evaluations
meta.policy	object	Policy evaluation result
meta.agent	object	Agent identity and fingerprint

№ 05The Case

Why OmegaEngine

Compare what you get with OmegaEngine vs building it yourself.

Time to production

5 minutes

3–6 months

Safety layers

12 (parallel)

1–2 (if any)

Cryptographic audit

✓

✗

Multi-model arbitration

✓

✗

Decision latency

sub-second (governs every call)

varies

OmegaIQ / Regret Score

✓

✗

Compliance controls

SOC2/HIPAA-aligned

DIY

Kill switch

One toggle, system-wide

Manual restart

Feature	OmegaEngine	Build Your Own

№ 06Integration

3 Lines to Production

Drop in the SDK. Every agent action is now judged, signed, and auditable.

agent.ts

import { OmegaClient } from "@omegaengine/sdk";

const omega = new OmegaClient({
  baseUrl: "https://omegaengine.ai",
  apiKey: process.env.OMEGA_API_KEY,
});

// Every agent action — judged in sub-second time, signed, auditable
const result = await omega.decide({
  scenario: task.description,
  domain: "finance",
  riskTolerance: "LOW",
  agentName: "payment-processor-v3",
});

if (result.decision?.recommendedAction === "ESCALATE") {
  console.log("Escalated:", result.decision.summary);
  console.log("Regret Score:", result.decision.regretScore);
  console.log("Audit Hash:", result.meta?.auditHash);
  console.log("Halo ID:", result.meta?.omegaHaloId);
}

Ship AI That You Can Trust

Your AI agents make thousands of decisions per hour. Make sure every single one has a judgment, a receipt, and a kill switch.

Start Free

Read the Docs

The AI Judgment OS

OmegaEngine Core

One API call · 12 safety layers · Signed receipts

0ms

P95 · SAFETY LAYERS

SAFETY LAYERS

RESPONSE FIELDS

RED-TEAM VECTORS

Start Free

API Docs

№ 02Live Demo

Try the Decision Pipeline

Click the brain to send a scenario through all 12 safety layers. Watch the neural network light up.

omega-decision-engine — neural pipeline

Agent Request

"Transfer $50,000 to unverified offshore account"

Click to send this scenario through all 12 safety layers

№ 03The Engine

What Happens In That One API Call

12 Safety Layers

Safety classification, policy engine, anomaly detection, cryptographic audit — all firing in parallel. Sub-second governed decisions (excluding the LLM call).

ALLOW / BLOCK / ESCALATE

Deterministic verdicts with risk scores, reasoning chains, and full audit trail. Zero ambiguity.

Cryptographic Provenance

Multi-Model Arbitration

Shadow-evaluate across LLMs simultaneously. Consensus scoring with configurable confidence thresholds.

OmegaIQ + Regret Score

Proprietary intelligence metrics. OmegaIQ rates decision quality. Regret Score predicts the future cost of the action.

Policy Engine

Declarative domain-aware policies with risk tolerance levels, automatic versioning, and hot-reloading.

№ 04Response Schema

Full Response Anatomy

40+ fields in every response. Verdict, reasoning, cryptographic proof, agent identity, and more.

Field	Type	Description
decision.verdict	string	ALLOW \| BLOCK \| ESCALATE
decision.riskScore	number	0.0 – 1.0 risk assessment
decision.omegaIQ	number	Proprietary decision quality score (0–100)
decision.regretScore	number	Predicted future cost of this action
decision.impactScore	number	Magnitude of downstream impact
decision.reasoning	string	Human-readable explanation chain
decision.usage	object	Token counts and cost ($USD)
meta.omegaHaloId	string	Unique decision fingerprint
meta.auditHash	string	SHA-256 content hash (tamper-evidence)
meta.hmacSignature	string	Cryptographic HMAC signature
meta.envelope	object	Zod-validated canonical receipt
meta.arbitrationShadow	array	Multi-model shadow evaluations
meta.policy	object	Policy evaluation result
meta.agent	object	Agent identity and fingerprint

№ 05The Case

Why OmegaEngine

Compare what you get with OmegaEngine vs building it yourself.

Time to production

5 minutes

3–6 months

Safety layers

12 (parallel)

1–2 (if any)

Cryptographic audit

✓

✗

Multi-model arbitration

✓

✗

Decision latency

sub-second (governs every call)

varies

OmegaIQ / Regret Score

✓

✗

Compliance controls

SOC2/HIPAA-aligned

DIY

Kill switch

One toggle, system-wide

Manual restart

Feature	OmegaEngine	Build Your Own

№ 06Integration

3 Lines to Production

Drop in the SDK. Every agent action is now judged, signed, and auditable.

agent.ts

import { OmegaClient } from "@omegaengine/sdk";

const omega = new OmegaClient({
  baseUrl: "https://omegaengine.ai",
  apiKey: process.env.OMEGA_API_KEY,
});

// Every agent action — judged in sub-second time, signed, auditable
const result = await omega.decide({
  scenario: task.description,
  domain: "finance",
  riskTolerance: "LOW",
  agentName: "payment-processor-v3",
});

if (result.decision?.recommendedAction === "ESCALATE") {
  console.log("Escalated:", result.decision.summary);
  console.log("Regret Score:", result.decision.regretScore);
  console.log("Audit Hash:", result.meta?.auditHash);
  console.log("Halo ID:", result.meta?.omegaHaloId);
}

Ship AI That You Can Trust

Your AI agents make thousands of decisions per hour. Make sure every single one has a judgment, a receipt, and a kill switch.

Start Free

Read the Docs