Agent Store is now live — install your first enterprise agent in one click. Explore →
Platform Cloud Hybrid Enterprise Six Engines Orchestration Observability Harness Metaprise LLM
Module 03 · Execution Runtime Layer

Six engines.
One execution layer.

Six production-grade engines form the execution layer of the Agent OS. Each engine is configurable at runtime through the Dashboard — no code changes required. Together, they ensure every agent action is verified, compliant, executed, audited, stateful, and grounded in structured knowledge.

Enterprise AI demands more than inference.

LLMs generate text. Engines make it trustworthy. Every agent action passes through the Six Engines before, during, and after execution — ensuring that outputs are factually accurate, regulation-compliant, reliably executed, cryptographically audited, business-process-aware, and grounded in organizational knowledge.

6
Production Engines
0
Code Changes to Configure
100%
Actions Pass Through Engines
3
Deployment Modes Supported
Engine 01

Truth Engine

Fact verification and hallucination detection. Every claim an agent makes is checked against authoritative sources before it reaches your users or downstream systems.

Financial Reports Legal Research Medical Summaries Customer-Facing AI
Core Capabilities
Source Verification: Cross-references agent outputs against authoritative data sources, internal documents, and knowledge bases in real time
Hallucination Detection: Identifies fabricated facts, invented citations, and unsupported claims before they propagate downstream
Confidence Scoring: Assigns verifiable confidence scores to each factual claim — agents that cannot verify a statement flag it rather than assert it
Citation Tracing: Every verified fact links back to its source document, database record, or knowledge graph node for full auditability
How It Works
Runs inline with every agent inference pass — zero additional latency on cached knowledge
Configurable strictness levels: Advisory (log only), Warn (flag to user), Block (prevent output)
Integrates with the Knowledge Graph Engine for structured fact retrieval
Engine 02

Compliance Engine

Real-time compliance enforcement. Regulatory rules are applied deterministically — not by model judgment — to every agent action, every output, every decision.

HIPAA FINRA SOX GDPR FedRAMP
Core Capabilities
Rule-Based Enforcement: Compliance rules are deterministic code, not LLM-based judgment — PERMIT, DENY, or ESCALATE decisions are predictable and reproducible
Pre-Execution Screening: Every agent action is checked against the active compliance ruleset before execution begins
PII/PHI Detection: Automatic detection and redaction of personally identifiable information and protected health information in agent inputs and outputs
Regulatory Framework Templates: Pre-built rulesets for HIPAA, FINRA, SOX, GDPR, and FedRAMP — ready to activate, customizable per organization
How It Works
Integrated with AURA's Policy Engine for organizational authority enforcement (L1–L4 approval levels)
Rules configured via the Dashboard — compliance teams can update policy without engineering involvement
Every compliance decision is logged to the Audit Engine with full decision reasoning
Engine 03

Execution Engine

Task execution and tool-call management. The core runtime that coordinates agent actions, manages tool invocations, and ensures reliable completion of every mission step.

Multi-Step Workflows API Orchestration Data Pipelines System Integration
Core Capabilities
Tool-Call Gateway: Manages all agent-to-tool communication with DualToken authentication — every external call is authorized and scoped
Durable Execution: Crash recovery and event replay ensure tasks are never lost, even during infrastructure failures
Saga Compensation: Compensating transaction pattern for automatic rollback on multi-step failures — partial execution never leaves your systems in an inconsistent state
Checkpoint Persistence: Per-step state checkpoint persistence — long-running missions resume from the last successful step, not from the beginning
Reliability Features
Retry Policy Engine: Configurable retry policies with exponential backoff, max attempts, and non-retryable error classification
Activity Heartbeat: Progress monitoring for long-running tasks — stalled executions are detected and recovered automatically
Continue-As-New: State renewal for extremely long executions, overcoming single-execution history limits
Engine 04

Audit Engine

Tamper-proof execution records. Every agent action generates a cryptographic audit entry via AuditChain — a SHA-256 hash-chained log that cannot be altered after the fact.

Regulatory Audit Forensic Analysis Incident Response Board Reporting
Core Capabilities
AuditChain: SHA-256 hash-chained tamper-proof audit log — every entry links cryptographically to the previous entry, making retroactive alteration detectable
Synchronous Write: Audit entries are written synchronously with every agent action — the audit trail is never behind the execution
Complete Decision Trace: Records not just what happened, but why — including model inputs, confidence scores, policy decisions, and tool-call parameters
Export Anytime: Full audit history exportable at any time in standard formats — nothing is hidden, delayed, or aggregated before export
Compliance Integration
Satisfies SOC 2 Type II logging requirements across all five trust service criteria
HIPAA audit trail requirements met natively — no additional tooling needed
Hybrid and Enterprise deployments store AuditChain inside your perimeter — audit data never leaves your network
Engine 05

Business State Machine

Business process state management. Maps AI agent workflows to formal business process states — ensuring agents follow your organization's actual operational logic, not just LLM guesswork.

Loan Origination Claims Processing Order Fulfillment Approval Chains
Core Capabilities
Formal State Definitions: Define business process states explicitly — agents can only transition between valid states, never skip steps or invent shortcuts
Transition Guards: Each state transition has configurable guard conditions — the engine blocks transitions that violate business rules regardless of what the LLM suggests
MissionState Integration: Integrates with AURA's 8-state Mission lifecycle (CREATED → PLANNING → WAITING_APPROVAL → APPROVED → PENDING_EXECUTION → IN_PROGRESS → EXECUTED → CLOSED)
Process Visualization: Real-time visibility into where every active mission sits in your business process — via Dashboard or API
Why It Matters
Agents can reason creatively within each step, but cannot deviate from your defined business flow
Every state transition is recorded in the Audit Engine — complete process compliance history
Business analysts configure processes in the Dashboard — no engineering required
Engine 06

Knowledge Graph Engine

Structured knowledge storage and retrieval. Gives agents access to your organization's structured knowledge — entities, relationships, hierarchies, and domain ontologies — not just unstructured text embeddings.

Regulatory Knowledge Product Catalogs Org Hierarchies Medical Ontologies
Core Capabilities
Entity-Relationship Store: Store structured knowledge as typed entities and relationships — not just vector embeddings of flat documents
Graph Queries: Agents traverse multi-hop relationships to answer complex questions that require connecting information across domains
Domain Ontologies: Define industry-specific ontologies (financial instruments, medical terminology, legal citations) that agents use for precise reasoning
Incremental Updates: Knowledge graph updates in real time as new information arrives — agents always work with current organizational knowledge
Integration With Other Engines
The Truth Engine queries the Knowledge Graph for source-of-truth verification
The Compliance Engine references regulatory knowledge stored as structured graph data
The Business State Machine uses entity relationships to determine valid state transitions

How the six engines work together.

Every agent action flows through the engines in a coordinated pipeline. The sequence ensures that by the time an action reaches your systems, it has been verified, authorized, and recorded — with full rollback capability if anything fails.

Step 01
Knowledge Graph
Retrieves structured context and domain knowledge
Step 02
Truth Engine
Verifies facts and detects hallucinations
Step 03
Compliance
Enforces regulatory rules and policies
Step 04
State Machine
Validates business process state transition
Step 05
Execution
Runs the action with durable guarantees
Step 06
Audit
Records the complete chain to AuditChain

Runtime configuration guarantee: All six engines are configurable at runtime through the Dashboard. Compliance teams, business analysts, and security officers can adjust engine behavior without engineering involvement or code deployments.

See the engines in action.

Deploy your first agent today. Every plan includes all six engines — no capability tiers.