Enterprise AI Agents · Software In A Service

AI agents that ship work, governed like employees, not chatbots.

Production-ready agents engineered around your processes: 506 catalogued agents across 8 domains, each a first-class principal with its own identity, scoped credentials, and full action logging. Designed, secured, hosted, and monitored by Citadel.

Book a discovery call Download the AI Agent Blueprint

8Domains
33Function groups
506Catalogued agents
3Autonomy tiers

A governed AI agent catalog is a curated set of production agents that each hold their own identity rather than sharing a service account, and Citadel Cloud Management publishes one as Software In A Service, or SIAS: 506 catalogued agents across 8 domains, designed, secured, hosted, and monitored by Citadel. Every agent is a first-class principal with scoped credentials and full action logging, and risk is bounded by setting autonomy per agent and per action across three tiers. Tier 1 is read-and-report: triage, analysis, and drafting run unattended, because the agent observes and recommends but changes nothing. Tier 2 is supervised: writes to systems of record happen within scoped credentials and change windows, and every write is attributable and reversible. Tier 3 is human-gated: anything touching money, people, or legal is draft-only behind named approvers, with four-eyes enforcement. Autonomy is graduated deliberately: discovery, agent selection and grounding, a sandboxed rollout at Tier 1, then monitored operation.

Most “AI agents” are a demo that impresses in a meeting and stalls in production: no identity, no guardrails, no audit trail, no owner. The blocker to real automation isn’t the model. It’s governance: who is allowed to do what, with which credentials, and how you prove it afterward.

One governance model, applied to every agent

Every agent is a first-class principal with its own identity and full action logging. Autonomy is set per agent, per action, across three tiers, so you graduate trust deliberately instead of betting the business on a black box.

Autonomous

Read-and-report. Triage, analysis, and drafting run unattended: the agent observes and recommends, but changes nothing.

Supervised

Writes to systems of record within scoped credentials and change windows. Every write is attributable and reversible.

Human-gated

Anything touching money, people, or legal is draft-only behind named approvers, with four-eyes enforcement.

Every action takes the same path, and the tier is set per agent and per action. T1 observes and changes nothing; T2 writes to systems of record inside change windows, attributably and reversibly; T3. Anything touching money, people, or legal. Stops at a named approver before it happens.

Delivery: Discovery → agent selection & grounding → sandboxed rollout at T1 → graduated autonomy → monitored operation.

What these agents actually change

The catalog is deep, but the buying question is simple: which business problem gets cheaper, faster, or safer, and at which autonomy tier the work runs.

Reduce support & operations cost

Triage, classification, and first-draft responses run continuously so specialists spend their hours on the exceptions, not the queue.

Mostly T1 to T2

Accelerate software delivery

Code review, test scaffolding, dependency and release chores handled by governed agents, shorter cycle time without loosening the merge bar.

T1 to T2

Strengthen security operations

Alert triage and evidence collection at T1, containment left human-gated at T3. Analysts stop drowning in noise while control of action stays with people.

T1 + T3 gate

Automate compliance evidence

Control checks, evidence gathering, and audit-trail assembly run on a schedule and are attributable per agent. Audits become a query, not a fire drill.

T1 to T2

Lower cloud spend

Cost-anomaly detection, rightsizing recommendations, and idle-resource reporting surface savings; the changes stay supervised or human-gated.

T1, writes T2

Increase engineering velocity

Runbook execution, ticket enrichment, and knowledge lookups offloaded from senior engineers, the same team ships more without more headcount.

T1 to T2

The catalog: 8 domains, 506 agents

Agents are chosen from this catalog and grounded in your corpora: runbooks, ADRs, policies, catalogs. Search for a capability, tool, or domain to see what fits your stack.

Search the catalog

506 agents across 8 domains

Build, ship, and run software and infrastructure

6 function groups · 96 agents

Software Engineering

The full development lifecycle from design through maintenance, integrated with repos, CI, and issue trackers.

DevOps and Platform Engineering

Delivery pipelines, runtime operations, reliability, and cost. Incident-facing agents pull from live telemetry through read-scoped tools.

Network Operations

Configuration, connectivity, and traffic engineering across on-prem, cloud, and edge. Config-writing agents are T2 with change-window enforcement.

Cloud and Enterprise Architecture

Design-time agents: blueprints, reviews, and migration planning grounded in the organization's own standards and reference architectures.

Kubernetes and Platform Engineering

Cluster lifecycle, workload policy, and the internal developer platform. Admission-policy and upgrade agents are T2 with change-window enforcement; cluster-mutating actions are human-gated.

Cluster Architecture
Upgrade and Version Skew Planning
Helm Chart Generation
Admission Policy (Kyverno / Gatekeeper)
Kubernetes Cost and Right-sizing
Cluster Security Hardening (CIS / PSS)
Service Mesh Operations
GitOps Reconciliation
Internal Developer Portal
Managed Kubernetes Reliability (EKS / AKS / GKE)

Cloud Platform Operations

Provider-native control planes, where the operating model genuinely differs per cloud. These sit alongside the vendor-neutral CSPM and FinOps agents rather than replacing them.

Scope build agents, book a call

Defend, detect, govern, and prove it

3 function groups · 56 agents

Security Operations

Around-the-clock triage, hunting, and response. Tier-1 triage agents run at T1; anything that isolates a host or disables an account requires a human gate.

SOC Analyst (Tier-1 Triage)
Threat Hunting
Threat Intelligence
Detection Engineering (SIGMA / KQL / YARA)
Vulnerability Management
Penetration Test Assistant
Malware Analysis
Phishing Triage
SIEM Investigation
SOAR Automation
Incident Forensics
Ransomware Response
Insider Threat
Attack Surface Management
Red Team Emulation
Purple Team Exercise
IAM
Identity Governance
PAM Review
DLP
Secrets Leak Detection
Cloud Security
CSPM
CNAPP
Container and K8s Security
Supply Chain Security (SBOM)
AI / LLM Security
Zero Trust Advisor
Security Awareness Training
Risk Assessment
Threat Modeling
API Security Review
Dependency Vulnerability Triage

Governance, Risk and Audit

Frameworks into evidence, evidence into audits. These agents make compliance continuous instead of an annual scramble.

Compliance Mapping (SOC 2 / ISO 27001 / HIPAA / PCI / FedRAMP)
Policy Drafting
Control Testing
Evidence Collection
Internal Audit
Third-party / Vendor Risk
Enterprise Risk Register
Regulatory Change Monitoring
Privacy (GDPR / CCPA / DSAR)
Data Protection Impact Assessment
Business Continuity and DR Planning
AI Governance (Model Risk / EU AI Act)
ESG Reporting

Workforce Identity and Access

Human identity: joiners and leavers, entitlements, and the attack paths between them. Distinct from the agent-identity groups below, which govern non-human principals.

Identity Architecture
Identity Estate Discovery
Access Certification Campaign
Joiner-Mover-Leaver (Workforce)
Role Mining
RBAC Right-sizing
ABAC Policy Authoring
Identity Attack Path Analysis
Passwordless and Phishing-Resistant MFA Migration
Entra ID / Okta Tenant Security

Scope defend agents, book a call

The data estate and the machinery of intelligence

4 function groups · 68 agents

Data Engineering and Analytics

Pipelines, quality, governance, and the analytics surface. SQL-writing agents run against read-only replicas with row-level security intact.

AI, ML and LLMOps

The platform operating on itself: model lifecycle, agent quality, safety, and spend. These agents gate what every other agent is allowed to become.

AI Platform Engineering

Running the AI platform as infrastructure: serving, capacity, and promotion. Where AIM governs how a model behaves, these govern where it runs and how it ships.

AgentOps and Agent Lifecycle

The build-test-ship-retire loop for agents themselves. Retirement is a first-class step: a decommissioned agent whose credentials outlive it is the failure mode this group exists to prevent.

Agent Builder
Agent Test Harness
Agent Benchmarking
Agent Deployment and Promotion
Agent Version Control and Rollback
Agent Runtime Observability
Agent Security Scanning (Injection / Tool Abuse)
Agent Retirement and Decommissioning

Scope data estate and the machinery of intelligence agents, book a call

The functions that run the company

6 function groups · 105 agents

IT Service Management

The employee-facing service desk and the asset estate behind it. Deflection-first: resolve at T0-T1, escalate with full context.

Human Resources

Hire-to-retire. Screening and review agents carry bias-audit requirements in their eval suites as a deployment precondition.

Finance and Accounting

Transaction processing through planning. Anything that moves money is T3: draft-and-approve only, full audit trail, four-eyes enforcement.

Procurement and Supply Chain

Source-to-pay and plan-to-deliver. Supplier risk agents merge external signals with contract obligations from the legal corpus.

Legal

Contracts, counsel, and discovery. Uses the graph-walk retrieval profile against a clause knowledge graph; privileged corpora are tenancy-isolated.

Product, Project and Program Management

From product definition to portfolio delivery. Synthesis-heavy agents that turn meetings, feedback, and tickets into decisions and artifacts.

Scope functions that run the company agents, book a call

Find, win, keep, and grow customers

3 function groups · 59 agents

Sales and Revenue Operations

Prospect to renewal. CRM-writing agents are T2; pricing approval workflows stay human-owned.

Marketing and Communications

Demand, brand, and comms. Generation agents draft against the brand corpus; publishing is always a human action or an approved automation with rollback.

Customer Experience and Support

Every support channel plus the analytics behind them. Deflection with dignity: grounded answers, cited sources, clean handoff with full context when escalation is right.

Scope find agents, book a call

Specialized agents for regulated and domain-specific work

6 function groups · 72 agents

Healthcare and Life Sciences

Clinical, administrative, and research workflows. FHIR R4/R5-aware retrieval; clinical suggestion agents are decision support, never autonomous diagnosis.

Financial Services and Banking

Regulated money movement and market-facing work. Every decision-support output carries model-risk documentation per SR 11-7-style governance.

Insurance

First notice of loss through subrogation. Adjudication agents recommend; adjusters decide above defined thresholds.

Retail and E-commerce

Catalog to conversion. Pricing agents operate inside guardrail bands with margin floors enforced by policy, not prompt.

Manufacturing, Energy and Industrial

OT-adjacent agents are strictly read-only against historians and SCADA exports; nothing on this platform writes to a control system.

Public Sector and Defense

Cleared-environment patterns: IL-appropriate hosting, CUI handling, and FedRAMP-aligned controls from the GRC corpus. Air-gapped model options via the gateway.

FOIA Processing
Grants Management
RMF / ATO Documentation
Export Control (ITAR / EAR)
Citizen Services Assistant
Case Management
Acquisition Compliance (FAR / DFARS)
Benefits Eligibility Screening
FedRAMP Authorization
RMF Assessment
NIST 800-53 Control Mapping
POA&M Management
Continuous Monitoring (ConMon)
Security Control Assessor
Classified Environment Compliance (IL4 / IL5)
Zero Trust Federal Architecture

Scope specialized agents for regulated and domain-specific work agents, book a call

The widest lens, the strictest access control

1 function group · 18 agents

Executive Copilots and Strategy

Per-role copilots plus the strategy functions that serve the whole leadership team.

Scope widest lens agents, book a call

Every agent is a first-class principal

4 function groups · 32 agents

Agent Identity and Credentialing

First-class identity per agent instance. Attestation binds a running workload to its registry entry; every credential is short-lived and issued by the broker, never checked into code or config.

Delegation and Authorization

Every action an agent takes on a user's behalf carries a verifiable delegation chain: token exchange at the boundary, purpose binding on the grant, and scoped tool permissions evaluated at a policy decision point on every call.

Non-Human Identity Governance

Agents multiply non-human identities faster than quarterly reviews can catch. Continuous discovery, ownership attribution, rotation enforcement, and joiner-mover-leaver lifecycle applied to machines the same way it is applied to people.

Identity Threat Detection and Response

Runtime defense for agent identities: detect impersonation, token theft and replay, confused-deputy patterns, and privilege escalation, then revoke fast through kill switches wired into the model gateway.

Scope every agent is a first-class principal agents, book a call

Industry solutions

The same governance model, mapped to where the work is in your sector. Every rollout starts at Tier-1 read-and-report and graduates autonomy deliberately. Nothing here is a timeline or savings promise until we scope it with you. Full landing pages for fifteen industries are here.

Healthcare

Problem: Prior-auth, intake, and documentation backlogs; PHI that cannot leave your control.
Recommended agents: Support & operations, data & knowledge, compliance agents, grounded in your policies.
Expected outcome: Routine documentation and eligibility checks move faster while PHI stays tenancy-isolated and every action is logged.
Rollout: Start T1 read-and-report in a sandbox; graduate writes only after HIPAA-aligned review.

Healthcare agents in depth →

Financial Services

Problem: KYC/AML review queues, reconciliation, and audit trails under heavy regulatory load.
Recommended agents: Compliance, data & knowledge, and security-operations agents with four-eyes on any write.
Expected outcome: Analysts spend time on genuine exceptions; every decision path is attributable and reversible for examiners.
Rollout: T1 triage first; T2 reconciliation under scoped credentials; anything moving money stays T3.

Financial Services agents in depth →

Government & Public Sector

Problem: Case backlogs and FOIA/records workloads with strict provenance and access requirements.
Recommended agents: Data & knowledge and compliance agents, single-tenant, isolated per program.
Expected outcome: Faster case handling and records retrieval with a defensible, per-agent audit trail.
Rollout: Sandboxed T1 rollout; FedRAMP-aligned controls; deliberate autonomy graduation.

Government & Public Sector agents in depth →

Energy & Utilities

Problem: Asset monitoring, outage triage, and field-report processing across dispersed systems.
Recommended agents: Operations and data agents that watch telemetry and draft the response.
Expected outcome: Anomalies surface and get triaged sooner; operators keep control of every field action.
Rollout: T1 monitoring and reporting; T2 ticketing; physical/control actions human-gated.

Energy & Utilities agents in depth →

Manufacturing

Problem: Supply-chain exceptions, quality-report processing, and maintenance ticket load.
Recommended agents: Operations, data & knowledge agents grounded in your SOPs and catalogs.
Expected outcome: Exception handling and reporting speed up; scarce specialists focus on true escalations.
Rollout: T1 detection and drafting; T2 system-of-record updates in change windows.

Manufacturing agents in depth →

Retail & E-commerce

Problem: Support volume spikes, catalog and content upkeep, and returns processing.
Recommended agents: Support & operations and content agents that draft, classify, and route.
Expected outcome: Faster responses and cleaner catalogs during peaks without over-hiring seasonal staff.
Rollout: T1 drafting and classification; T2 catalog and ticket writes under supervision.

Retail & E-commerce agents in depth →

Insurance

Problem: Claims intake, document extraction, and policy-servicing queues.
Recommended agents: Data & knowledge, compliance, and support agents with reversible writes.
Expected outcome: Straight-through handling of routine claims; adjusters focus on complex or disputed cases.
Rollout: T1 extraction and triage; T2 servicing updates; payouts and denials stay T3.

Insurance agents in depth →

Technology & SaaS

Problem: Growing support tickets, on-call toil, and slower delivery as the product scales.
Recommended agents: Engineering, security-operations, and support agents across the delivery lifecycle.
Expected outcome: Support deflection and less on-call toil while engineers ship more per sprint.
Rollout: T1 triage and enrichment; T2 runbook execution; production changes gated.

Technology & SaaS agents in depth →

Enterprise packages

Fixed-fee engagements sized to where you are, from a single-workflow pilot to a governed platform you own. Priced after a discovery call, no hourly meter. See full pricing. The three-week Agent Assurance Assessment is priced at a fixed $15,000, the same anchor as the Starter Pilot floor, with its full scope and deliverables published on the page.

Starter Pilot

From $15K fixed-fee

Prove value on one high-friction workflow.

1 to 2 governed agents, Tier-1 read-and-report
Grounded in one corpus (runbooks, policies)
Sandboxed rollout with full action logging

Most chosen

Professional

$50K to $150K fixed-fee

Automate a whole function or domain.

A domain of agents at T1 to T2
Scoped credentials, change-window writes
SSO + audit integration, monitored operation

Enterprise

$150K to $500K+ fixed-fee

Multi-domain rollout across the org.

Agents across multiple of the 8 domains
Full T1 to T3 governance with named approvers
Security review + FedRAMP-aligned controls

Managed AI Workforce

From $8K per month

Run and improve the agents you deployed.

Ongoing operation, monitoring, and tuning
Autonomy graduated as trust builds
SLA-backed where the engagement calls for it

Custom AI Platform

Custom engagement

Own a governed agent platform end to end.

Your own multi-agent platform, IP yours
Identity, tiering, and audit built in
Runbooks + clean handoff to your team

Executive ROI calculator

A back-of-envelope from your own numbers. Hours a governed agent workforce can return, valued against your investment. It is an illustration, not a quote.

Illustrative first-year estimate

$143,520Annual labor value freed

Hours returned / year: 2,208
Net first-year benefit: $23,520
First-year ROI: +20%
Payback period: 11 mo

An illustration from your inputs, not a quote. Real savings depend on which agents fit, the autonomy tier, and your data. We size it precisely in a discovery call.

Pressure-test these numbers with an architect

Ways to start

Four low-commitment entry points. Most engagements begin with a free call or the readiness assessment, then a scoped pilot.

Free

Free 30-min strategy call

Scope the highest-value, lowest-risk agents for your stack. No pitch deck.

Book it →

From $1,500

Readiness Review (pre-assessment)

A short structured review of your data, workflows, and controls with a prioritized agent roadmap. Scopes whether the fixed-price Agent Assurance Assessment is the right next step.

Book it →

Half or full day

Architecture Workshop

A working session that leaves you with a governed reference architecture, not slides.

Book it →

Free

Executive Briefing

The Enterprise AI Agent Blueprint, the governance model behind SIAS, sent to your inbox.

Get the blueprint →

Where do the agents run and who holds our data?

Agents run in or adjacent to your tenancy and are grounded in your corpora, which stay under your control and are tenancy-isolated per engagement. The full posture is on the security & governance page.

Is the deployment single-tenant?

Yes. Corpora and agent identities are isolated per engagement, and privileged corpora are isolated per legal domain so one workstream cannot read another.

What if an agent hallucinates on a write?

Writes only happen at supervised (T2) or human-gated (T3) tiers within scoped credentials. Every action is logged, attributable to a specific agent identity, and designed to be reversible, so a wrong action is traceable and can be rolled back. Anything touching money, people, or legal is draft-only behind named approvers.

How are agent credentials managed?

Each agent is its own principal with attested workload identity (SPIFFE/SPIRE). Credentials are broker-issued, ephemeral, and purpose-bound through OAuth token-exchange (RFC 8693). There are no static long-lived keys.

How do we start without betting the business on it?

Engagements start with lower-risk Tier-1 read-and-report agents in a sandboxed rollout, then graduate autonomy deliberately as trust builds. You set the tier per agent, per action.

How is a SIAS engagement priced?

Fixed-fee, scoped after a discovery call, from a Starter pilot around $15K to multi-domain Enterprise rollouts, with an optional monthly Managed AI Workforce retainer. You know what a phase costs before you commit; there is no hourly meter. Detailed pricing is on the enterprise page.

Who owns the agents we build together?

The configuration and deliverables built for your environment are yours; specific IP terms are set in the engagement agreement so your legal team reviews them before signing.

Ready to see which agents fit your stack?

A 30-minute discovery call to scope the highest-value, lowest-risk agents first, usually T1 read-and-report, then graduate autonomy as trust builds. Or read the security & governance posture.

Book a discovery call Pick a time

One governance model, applied to every agent

Autonomous

Supervised

Human-gated

What these agents actually change

Reduce support & operations cost

Accelerate software delivery

Strengthen security operations

Automate compliance evidence

Lower cloud spend

Increase engineering velocity

The catalog: 8 domains, 506 agents

Build, ship, and run software and infrastructure

Software Engineering

DevOps and Platform Engineering

Network Operations

Cloud and Enterprise Architecture

Kubernetes and Platform Engineering

Cloud Platform Operations

Defend, detect, govern, and prove it

Security Operations

Governance, Risk and Audit

Workforce Identity and Access

The data estate and the machinery of intelligence

Data Engineering and Analytics

AI, ML and LLMOps

AI Platform Engineering

AgentOps and Agent Lifecycle

The functions that run the company

IT Service Management

Human Resources

Finance and Accounting

Procurement and Supply Chain

Legal

Product, Project and Program Management

Find, win, keep, and grow customers

Sales and Revenue Operations

Marketing and Communications

Customer Experience and Support

Specialized agents for regulated and domain-specific work

Healthcare and Life Sciences

Financial Services and Banking

Insurance

Retail and E-commerce

Manufacturing, Energy and Industrial

Public Sector and Defense

The widest lens, the strictest access control

Executive Copilots and Strategy

Every agent is a first-class principal

Agent Identity and Credentialing

Delegation and Authorization

Non-Human Identity Governance

Identity Threat Detection and Response

Industry solutions

Healthcare

Financial Services

Government & Public Sector

Energy & Utilities

Manufacturing

Retail & E-commerce

Insurance

Technology & SaaS

Enterprise packages

Starter Pilot

Professional

Enterprise

Managed AI Workforce

Custom AI Platform

Executive ROI calculator

Ways to start

Free 30-min strategy call

Readiness Review (pre-assessment)

Architecture Workshop

Executive Briefing

Governance & deployment questions

Ready to see which agents fit your stack?