SONGLINES CONTROL

One Governance Evidence Layer Across Your Existing AI Estate.

Deploy Songlines Control in the pattern that matches your architecture: ingest existing telemetry read-only, call it as a Policy Decision Point from MuleSoft or API gateways, embed it through lightweight SDKs, or use the full managed control plane where appropriate.

Explore Architecture Review Regulatory Evidence Explore Live Demos
ENTERPRISE INTEGRATION

Deployment & Data Boundaries

Choose the trust boundary that fits your enterprise. Cetus AI supports multiple deployment patterns to match your risk appetite and existing architecture.

Read-Only Evidence

Ingest approved logs read-only and produce cross-stack governance evidence. No runtime policy decision.

Best for: Organisations starting with visibility and disclosure readiness.
Recommended

PDP Sidecar

Run the Policy Decision Point inside your environment; use Cetus AI management in Australia.

Best for: Mature enterprises seeking control without a new transit layer.

Dedicated Instance

Deploy the complete platform in a dedicated customer tenant or private cloud environment.

Best for: Regulated workloads and strict isolation requirements.

Managed Control Plane

Route selected AI workloads through the full managed platform where central enforcement is appropriate.

Best for: Greenfield workloads and consolidated AI services.
RUNTIME CONTROL

Policy Decision Point

Evaluate AI-specific policy without replacing your API gateway. Songlines Control returns deterministic decisions that your existing orchestration layer can enforce.

  • Approve: Continue to the selected AI provider.
  • Deny: Stop the request and return policy reason.
  • Conditions: Redact PII, substitute models, or require human review.
// Example PDP Response
{
  "decision": "APPROVED_WITH_CONDITIONS",
  "conditions": [
    "PII_REDACTION_REQUIRED",
    "HUMAN_REVIEW_REQUIRED"
  ],
  "policy_refs": ["ENTERPRISE-HR-003"],
  "audit_id": "aud-2026-08-22-7a3f2c91"
}
ADM Register
Inventory of systems making automated decisions
Cryptographic Audit Chain
Tamper-evident record of all governance events
Cost & Sustainability
Attribution of AI compute, spend, and emissions
ASSURANCE

Evidence Fabric

Correlate source logs and policy decisions into one governance record. Songlines Control links request IDs to telemetry from Azure Monitor, AWS CloudTrail, SAP, and Salesforce.

Explore Regulatory Compliance Export →
Platform Architecture

One platform. Four layers of enterprise AI control.

Use the capabilities your organisation needs while preserving one governance record across policy, evidence, value and enterprise integration. Availability and coverage depend on the selected deployment, connected systems and configuration.

Layer 01

Govern and Enforce

Apply AI-specific authority and policy to workloads routed through configured control paths.

  • Policy Engine and Runtime Guardrails
  • Agent Identity & Access
  • PII Detection and Human Review
  • Budget Controls and Model Routing
Layer 02

Observe and Evidence

Capture activity, cost, risk and decision evidence from connected sources.

  • Economic Control and Workflow Economics
  • Observability and Audit & Compliance
  • Shadow AI and Incident Management
  • Model Risk and Sustainability
Layer 03

Prove and Decide

Value Control connects approved baselines, fully loaded cost, outcome evidence, attribution and finance validation before reporting risk-adjusted portfolio value.

  • Forecast, observed and attributed value states
  • Finance validation and evidence completeness
  • Risk and correction-cost treatment
  • Scale, Optimise, Review or Retire decision records

Forecasts and operational estimates are not presented as realised ROI. Finance and business validation remain required.

Review Value Control →
Layer 04

Integrate and Extend

Work with the systems the enterprise already owns rather than requiring replacement.

  • MuleSoft architecture and evidence-contract pattern
  • Azure and AWS telemetry
  • SAP and Salesforce evidence patterns
  • CloudZero and Finout cost-adapter patterns; APIs and SDKs

Provider connections require customer credentials, source mapping and validation for the selected environment.

Economic Control Layer

Connect supported telemetry to attribute AI spend by model, workflow, team and user

The Economic Overview dashboard attributes the telemetry received from configured sources by model, workflow, team and user. Cost calculations use the selected provider rates and available token data; completeness depends on source coverage, mapping and customer configuration.

  • Real-time MTD spend with month-on-month trend comparison
  • Token-consumption tracking for models represented in connected telemetry
  • Model-routing scenario comparison against a defined baseline
  • Average cost per request for telemetry received from configured sources
  • CSV export for reconciliation with cloud billing and finance reporting
Songlines Control® — Economic Overview Dashboard
Economic Overview · Real-time AI cost tracking and budget management
Workflow Economics

Step-level cost breakdown for mapped AI workflows

Workflow Economics uses mapped telemetry to show available step-level cost and latency data for a selected workflow. The resulting breakdown supports customer investigation of potential optimisation opportunities.

  • Per-step cost and latency breakdown where workflow telemetry is mapped
  • Total cost, token count, latency, and step count per run
  • Cost per Step and Latency per Step charts side by side
  • Step-by-step breakdown with model, cost, latency, and token detail
  • Identify the most expensive steps for targeted optimisation
Songlines Control® — Workflow Economics
Workflow Economics · Step-level cost and latency breakdown per workflow run
Model Routing Engine

Compare model-routing scenarios before changing a production workload

Model-routing scenarios compare a current workload with lower-cost or policy-preferred alternatives. Actual savings and operational outcomes depend on workload, quality, latency, provider, policy and implementation constraints.

  • Residency-aware routing — direct eligible requests to approved endpoints under configured policy
  • Cost Optimisation — route eligible tasks to approved lower-cost models when configured
  • High-Stakes Analysis — legal and financial workflows routed to premium models
  • Fallback rules — route eligible requests to an approved alternative when a configured threshold is reached
  • Live routing decisions feed with model, latency, and policy enforcement status
Songlines Control® — Model Routing Engine
Model Routing Engine · Dynamic model selection rules and live routing decisions
Observability

Operational visibility for connected AI workloads — latency, errors and throughput

The Observability module presents latency, error, success and token-throughput telemetry received from connected sources. Attribution and completeness depend on the identifiers and fields supplied by each configured workload.

  • Request traces for telemetry received from configured workloads and control paths
  • Latency percentile charts (p50, p95, p99) over 24 hours
  • Error rate tracking with policy decision visibility (allowed / modified / blocked)
  • Success rate and average latency KPIs updated in real time
  • Per-model and per-workflow drill-down for root cause analysis
Songlines Control® — Observability
Observability · Real-time request traces, latency percentiles, and error monitoring
Audit & Compliance

Tamper-evident audit records designed to support Australian review requirements

Requests routed through configured control paths can generate signed, append-only records containing the fields available from the workload and policy decision. These records support audit and compliance review when combined with the customer's legal, risk, retention and operating controls.

  • Timestamped, tamper-evident records with configured retention and privileged administration
  • Full-text search across all audit fields — model, user, workflow, cost range
  • Date range filtering for regulatory submissions and legal discovery
  • CSV export to support customer review against Privacy Act, APS AI Policy and ISO 42001 requirements
  • Aligned with Privacy Act 1988, APS AI Policy, and ISO/IEC 42001
Songlines Control® — Audit and Compliance
Audit & Compliance · Tamper-evident records for covered AI interactions
Simulation Engine

Model what-if scenarios before committing to a routing change

The Simulation Engine models the potential cost and performance impact of routing strategy changes before production use. Results depend on the available telemetry, selected assumptions and customer review of risk, latency and quality constraints.

  • What-if scenario cards with current vs projected cost comparison
  • Risk ratings (Low / Medium / High) for each proposed routing change
  • Latency impact and quality impact projections per scenario
  • Modelled cost scenarios across selected routing strategies and assumptions
  • Apply Scenario button — changes only go live with explicit administrator approval
Songlines Control® — Simulation Engine
Simulation Engine · What-if scenarios and 6-month cost forecasting
User Management

Role-based access control with per-user AI spend attribution

User Management presents identity, usage and cost data received for registered users and connected workloads. Administrators can assign roles, review attributed activity and suspend platform accounts under the customer's access process.

  • Per-user AI request count, tokens used, and total spend in real time
  • Admin and user role assignment with server-side enforcement
  • Account suspension with immediate access revocation
  • CSV export for finance reporting and cost allocation by team
  • Last sign-in tracking for access review and compliance audits
Songlines Control® — User Management
User Management · Roles, access, and AI spending attribution per user
Budget Alerts

Guardrails that fire before you overspend — not after the invoice arrives

Configure alert rules scoped to model, workflow, team, or organisation-wide with custom dollar thresholds. Deliver alerts via HMAC-signed webhooks to PagerDuty, Slack, or Teams — or via email to your finance team. Cooldown windows prevent alert fatigue once a threshold is crossed.

  • Configurable rules with monthly, daily, or hourly rolling windows
  • Multi-channel delivery — webhook (HMAC-signed) and email
  • Cooldown enforcement to prevent duplicate alerts during active incidents
  • Full delivery log with HTTP status, response body, and timing
  • non-blocking alert latency on breach — fires on every telemetry ingest event
Songlines Control® — Budget Alerts
Budget Alerts · Configurable thresholds and real-time breach notifications
Governance Settings

Sovereign Mode, PII Auto-Redaction, and HITL Approval — all in one place

The Settings screen configures residency-aware routing, PII detection and redaction, and human-review gates for the workloads covered by the selected enforcement path. Customers remain responsible for classification rules, route coverage, exceptions and connected-provider configuration.

  • Residency-aware routing — apply approved endpoint rules to covered workloads
  • HITL Approval Workflows — human approval required for high-risk AI operations
  • PII Detection and Redaction — apply configured entity rules on covered enforcement paths
  • Prompt Injection Prevention — block prompt injection attempts in real time
  • Budget Threshold Alerts and notification preferences per user
Songlines Control® — Governance Settings
Governance Settings · Sovereign Mode, PII Auto-Redaction, and HITL controls
Agent Identity & Access Management New

The identity layer for your AI agent fleet

Organisations are deploying copilots, orchestrators, autonomous agents and third-party integrations across multiple environments. Agent IAM provides identity, authentication and authorisation controls for agents registered within the configured Songlines Control boundary.

  • Agent Identity Registry — register and lifecycle-manage agents within the configured governance boundary
  • Permission Boundaries — define the tools, data and APIs permitted for each registered agent
  • Credential Vaulting — configurable rotation schedules, revocation and governance records
  • Session Control — time-bound sessions with configured privilege de-escalation on expiry
  • Agent-to-Agent Trust — prevent unauthorised delegation chains between agents
  • Risk Scoring & Attestation — continuous risk evaluation based on permissions scope and action history
Songlines Control® — Agent Identity & Access Management Dashboard
Agent IAM · Register, authenticate, and govern every AI agent
Guardrails

Configurable controls across the AI request and response path

Requests routed through a configured Songlines Control enforcement path can be evaluated against applicable policy before provider execution, with response checks applied where configured. Customers remain responsible for route coverage, direct-provider access and exception controls.

Financial Guardrails

Configure spend thresholds by organisation, team, workflow or user. Covered requests can be blocked or routed to an approved alternative when a configured threshold is reached; effectiveness depends on route coverage and policy configuration.

  • Hard spend caps with automatic request blocking on breach
  • Soft threshold alerts via webhook and email before breach
  • Per-model, per-workflow, and per-user budget scoping
  • Hourly, daily, and monthly rolling windows

Data Sovereignty Guardrails

Residency-aware routing can restrict covered requests to approved endpoints and block a route that conflicts with the configured boundary. The effective residency position depends on the deployment pattern, connected providers, data flows and customer configuration.

  • Block endpoints outside the approved policy set on configured control paths
  • Per-workflow data residency rules (e.g., HR data stays onshore)
  • Record residency-policy outcomes for requests received through configured paths
  • Isolated deployment patterns for sensitive environments, subject to customer assessment

Privacy & PII Guardrails

Configured detection and redaction rules can identify selected personal-information types on covered outbound paths before provider submission. Customers remain responsible for classification quality, coverage, exceptions and validation in their environment.

  • Configured detection and redaction for selected PII entity types
  • Redaction applied on the configured enforcement path before provider submission
  • Redaction events recorded in the tamper-evident governance log
  • Configurable sensitivity levels per workflow or data classification

Model Access Guardrails

Define approved model policies for teams, workflows or users on covered control paths. Direct-provider access and unmanaged tools require complementary customer controls; Songlines Control does not claim universal coverage outside the configured architecture.

  • Per-team and per-workflow model allowlists and blocklists
  • Block unapproved model calls before they reach the provider
  • Reduce unmanaged model use by combining configured routing with customer access controls
  • Model version pinning to prevent silent capability changes

Prompt Integrity Guardrails

Configured prompt-integrity rules can flag, redact or block recognised adversarial patterns on covered inbound requests. Detection is one control within a broader application-security and model-risk approach.

  • Prompt-integrity checks on inbound requests routed through configured controls
  • Configurable sensitivity — warn, redact, or block on detection
  • Detected events logged with the context permitted by the customer's data-minimisation settings
  • Protects agentic workflows from adversarial user inputs

Human-in-the-Loop Guardrails

HITL Approval Workflows require a designated human approver to review and authorise high-risk AI operations before they are executed. Define which workflows, models, or request types require approval, set escalation paths, and maintain a full approval audit trail — ensuring human oversight is embedded in your AI governance framework, not bolted on afterwards.

  • Configurable approval gates per workflow, model, or risk classification
  • Approval requests routed via email, Slack, or webhook
  • Full approval audit trail — who approved, when, and why
  • Timeout and escalation rules for time-sensitive operations

Guardrails applied on the configured enforcement path

Requests routed through the configured enforcement path are evaluated against applicable policy before provider execution. Customers remain responsible for ensuring route coverage, limiting direct-provider access and managing authorised exceptions.

Integration

Connect a representative AI workload without replacing your existing stack

A lightweight telemetry pattern can be added alongside an existing AI response path. Implementation scope and timing depend on source quality, identity mapping, architecture and customer validation.

1. Create an API Key

Generate an API key scoped to the approved application, workflow or team under the customer's access-control process.

2. Wrap Your AI Call

Add a POST to /api/ingest after an AI response and map the available model, token, latency, cost and workflow fields.

3. Watch the Dashboard

Validate the received telemetry, attribution fields and cost assumptions before using dashboard outputs in governance or finance review.

# Add after every AI call — no changes to your existing AI code import requests, time # Your existing AI call (unchanged) start = time.time() response = openai.chat.completions.create(model="gpt-4o", messages=messages) elapsed_ms = int((time.time() - start) * 1000) # Instrument with Songlines Control® — add this single block requests.post("https://your-instance/api/ingest", headers={"Authorization": "Bearer sk-your-key"}, json={ "model": "gpt-4o", "input_tokens": response.usage.prompt_tokens, "output_tokens": response.usage.completion_tokens, "latency_ms": elapsed_ms, "workflow_id": "customer-support" })

TypeScript and Python SDK patterns are available. Provider and workload compatibility must be validated against the selected architecture and source data.

Deployment

Flexible deployment — from SaaS to air-gapped on-premises

Deploy Songlines Control® in the model that matches your organisation's security classification, data residency requirements, and procurement constraints.

Option 01 · SaaS

Microsoft Azure Cloud

Hosted in Microsoft Azure cloud and managed by Cetus AI. The selected Azure region, data flows, subprocessors and shared operational responsibilities are confirmed during architecture and security review.

Implementation: Scoped after source and workflow discovery
Best for: Managed SaaS deployment pathway
Residency: Confirmed through Azure deployment design and provider scope

Option 02 · Private Cloud

Your Tenancy

Deployed into a supported customer cloud tenancy. Provider, region, service compatibility, isolation boundaries and operating responsibilities are validated during architecture and security design.

Implementation: Scoped after architecture and security discovery
Best for: Enterprise with existing cloud investment
Residency: Determined by the selected region, services and configuration

Option 03 · Air-Gapped

On-Premises

Customer-hosted and isolated deployment patterns can support workloads with elevated security requirements, subject to architecture validation and the customer's security assessment.

Implementation: Scoped after architecture, dependency and support discovery
Best for: Sensitive or isolated workloads
Isolation: Subject to customer requirements and validated design

Regulatory Alignment

Purpose-built for Australian regulatory requirements

Songlines Control® is designed from day one for the frameworks that matter to Australian enterprise and government — not bolted on after the fact.

Privacy Act 1988

Supports customer privacy controls with configured PII detection and redaction, residency-aware routing and governance records for covered interactions.

APS AI Policy

Supports APS transparency, accountability and human-oversight processes through configurable evidence and review workflows.

ISO/IEC 42001

Supports AI Management System certification with documented controls, risk records, and audit evidence exportable in a single click.

Essential Eight

ACSC mitigation strategies supported through application control, patch management visibility, and privileged access management via RBAC.

ISM Controls

Provides technical controls and evidence that can support customer assessment against applicable ISM requirements; coverage depends on the deployed architecture and operating controls.

Data Sovereignty

Australian deployment options and customer-hosted patterns can support data-residency requirements. Boundaries depend on the selected deployment, connected providers, data flows and customer configuration.

AASB S2 & Climate Disclosure

Produces AI-emissions estimates and reporting inputs that can support a customer's climate-reporting process, subject to methodology, materiality, data quality and assurance review.

APS AI Policy 2024

Supports AI inventory, risk assessment and sanctioned-use workflows that can help agencies operationalise applicable policy requirements within their governance model.

Privacy Act 1988 APS AI Policy 2024 AASB S2 ISO/IEC 42001 Essential Eight ISM Controls Assurance Roadmap Corporations Act 2001 Review Evidence Support Australian Deployment Options
Sustainable AI

AI Governance and Emissions Reduction Are the Same Mechanism.

Purchased AI services may contribute to an organisation's Scope 3 inventory depending on its reporting boundary and methodology. Songlines Control® provides usage-based estimates and reporting inputs for customer review; classification, materiality and assurance remain the customer's responsibility.

Observe Mode
AI Emissions Baseline Inputs

Covered interactions can be attributed by team and model and mapped to a CO₂e estimate. The resulting data can support customer reporting and review, subject to methodology, materiality, source completeness and assurance requirements.

Enforce Mode
Modelled Routing Scenarios

Model-routing scenarios compare eligible alternatives using selected cost, quality, latency and policy assumptions. Realised financial and emissions outcomes require customer measurement and validation.

Orchestrate Mode
Reduced Provider Inference for Cache Hits

Where a validated cache response avoids a provider call, provider inference and associated usage charges may be reduced. Measurement boundaries and infrastructure overhead must be defined.

Explore Sustainable AI → Download White Paper
Engagement Pathways

Start with the level of governance you need today

Every organisation begins from a different point. Start with an assessment, validate value through a focused pilot, or design an enterprise deployment around your existing architecture.

Understand

Assess Your Starting Point

Identify governance gaps, Shadow AI exposure, evidence readiness, and the immediate controls your organisation should prioritise.

Take the Gap Assessment →
Validate

Run a Focused Pilot

Instrument a representative workload, validate visibility and evidence outputs, and build a practical implementation roadmap.

Explore Professional Services →
Scale

Design Enterprise Deployment

Define deployment boundaries, integrations, policy decision points, evidence requirements, and rollout sequencing for your environment.

Book an Architecture Review →

"Your AI systems are already running. The question is whether you're in control of them."

Review Songlines Control® against your environment

Scope a representative workload, confirm evidence and control requirements, and define delivery timing after source, security and architecture discovery.

Discuss an Engagement → Calculate Your ROI Email Us
Interactive demonstrations · External Songlines Control experience

Explore Songlines Control

Choose the experience that best matches the question you want to explore. Each demonstration opens as a standalone Songlines Control experience in a new browser tab; it is not embedded within cetusai.com.au.

Guided Simulator

Create a 30-day synthetic operating environment and follow an illustrative journey from AI activity through governance evidence and value review.

Open Guided Simulator — new tab

Live Value Control Demo

Explore synthetic portfolio records across Forecast, Observed, Attributed, Finance validated and Risk adjusted states.

Finance validation remains an authorised Finance decision. Demonstrated attribution does not independently establish causation or realised ROI.

Open Value Control — new tab

Narrated Product Tour

Take a guided 15-step, approximately seven-and-a-half-minute walkthrough from AI operations to accountable value decisions, with audio controls.

Open Narrated Tour — new tab

Demonstration boundaries

All organisations, identities, workflows, events, costs, benefits, incidents, decisions and outputs shown are synthetic and illustrative. FinOps evidence is simulated, local to the demonstration environment and read-only; no external Finance or FinOps account is connected or changed. Benchmarks are directional and source-labelled, not predictions or financial advice.

Value Control keeps observed change, attribution, Finance validation and risk adjustment distinct. These demonstrations do not independently establish causation, realised ROI, compliance or assurance.

Demo access updated September 2026. External experiences may use session cookies and security controls; no iframe is used.

Platform Deep Dive · Release 2026.09.5

Explore the Cetus AI Platform

Follow a connected walkthrough across Teams, Control, Evidence Fabric and the integrated Value Control module. See how configured workloads can be governed, observed and reviewed without replacing the enterprise systems already in place. All demonstration records are synthetic and illustrative.

17:23 · Updated September 2026 · Australian English captions available