Itential

Itential names its integrations by class and vendor: ITSM (ServiceNow, Jira), infrastructure as code (Ansible, Terraform, OpenTofu), CI/CD (GitHub, Jenkins), observability (Datadog, Splunk), AIOps (Selector, Forward Networks), and security and vault systems, alongside cloud providers (AWS, Azure) and multi vendor network estates (Cisco, Juniper, Arista, Palo Alto, VMware).

Action is documented, not only a catalog. The Itential Gateway is a named execution engine that runs Python scripts, Ansible playbooks, OpenTofu and Terraform plans, and AI initiated actions directly against the customer's infrastructure with secure multi vendor connectivity, an authenticated action inside systems and devices Itential does not own.

The ServiceNow App, published in the ServiceNow App Store, pushes governed Itential automations into the customer's ServiceNow catalog as native Flow steps, so Itential's actions run inside another vendor's product as first class objects. Existing Ansible, Python and OpenTofu assets become governed services without rewriting.

The published counts differ: the architecture page says the platform unifies 300+ integrations into a single orchestration fabric, while the pricing and platform pages describe 1,000+ open source integrations in the Itential Marketplace. They are plausibly scoped differently, governed prebuilt connectors against the open community library, but as published they read as two counts of the same thing.

Sourceitential.com/cloud-platform/architecture, /pricing and /marketplaceread 2026-09-08

Two execution modes run side by side, and both are priced. The Workflow Engine runs pre-defined logic identically every time at a published throughput, 10 tasks per second started on Scale, 50 on Enterprise and no limit on Custom, while FlowAgents combine LLM reasoning with governed tools to solve a goal described in natural language, metered in agent sessions.

Workflow Studio is the authoring surface, a drag and drop builder with versioning and testing requiring no scripting, and Lifecycle Manager orchestrates "multi-day sequences with explicit state and full rollback" across the resource lifecycle from provisioning to retirement.

The execution involves many participants across domains.

Workflows are published as microservices through an open API to northbound systems and AI agents, so a change can be initiated by a person through the Operations Manager self service portal, an ITSM ticket in ServiceNow, a CI/CD pipeline, an AIOps detection or an external LLM through MCP, and the same governed path executes it across network, cloud, security and IT infrastructure.

Stateful orchestration tracks services with resource models that detect drift and automate lifecycle changes. This is a deterministic orchestration engine with a reasoning layer on top.

Sourceitential.com/pricing and /cloud-platform/architectureread 2026-09-08

Agents are grounded in a live, federated view of the customer's systems rather than a stored corpus.

Federation and Unified Data Models create a unified view of resources across disparate systems so agents and workflows operate on normalized data models, with Data Transformation converting diverse formats into common JSON that people and agents can interpret across multi vendor environments.

The FlowMCP Gateway lets agents invoke external MCP tools, NetBox, Selector and Forward Networks, for intelligence enrichment during execution, and Configuration Manager adds golden templates and compliance plans as a stored reference state.

The absence of a stored corpus is by design, and Itential says so: "Itential does not create copies of data, allowing your systems of record to serve as the source of truth." There is no maintained index, graph or embeddings layer over the customer's knowledge because the platform is built not to hold one; it federates a live view and enriches per execution, which is context assembled per run.

Sourceitential.com/cloud-platform/architecture and /cloud-platform/flowairead 2026-09-08

A review gate ships as a product line on every pricing tier: Work Center, a "human-in-the-loop task queue for operators" that handles approvals and inputs when agents and workflows need a person. Around it sit approval gates, blast radius limits, pre and post validation checks and rollback, which Itential states apply "equally to AI-triggered and human-initiated changes".

The architecture makes the boundary structural: "AI never touches infrastructure directly." Agents observe state, analyze conditions and propose actions, and every change flows through the orchestration engine with policy enforcement; agent proposed automations pass validation, testing and approval before production deployment.

Autonomy is a customer set dial rather than a vendor posture, from human in the loop to human on the loop, with a documented read only first adoption path. That is a review and approve surface over what the agent does, plus a reversal route, plus a configurable boundary, alongside RBAC.

Sourceitential.com/cloud-platform/architecture and /pricingread 2026-09-08

Security covers both attestation and access. On attestation, the architecture page's compliance section invites buyers to "verify Itential's security, availability, confidentiality, and processing integrity with SOC2", naming four Trust Services Criteria, and the pricing matrix names SOC 2 Type II in the Security and Governance line on all five tiers.

Privacy is stated as GDPR and CCPA compliance, and a Trust Center is published at itential.com/trust-center, linked from the Company menu and the site footer. Access runs through single sign on (SSO) via OpenID Connect or SAML, with programmatic identity for AI agents, RBAC and GBAC, granular permissions, secrets management, approval workflows for all infrastructure changes, and encryption in transit and at rest.

The governance model is explicitly identity aware for non human actors: RBAC policies apply consistently to human users and AI agents, the MCP Server authenticates by user or agent identity, and platform policy determines execution eligibility per caller. An immutable audit trail and on premises or customer controlled deployment back this up.

Sourceitential.com/cloud-platform/architecture, /pricing and /trust-centerread 2026-09-08

The audit record is scoped to the agent. Audit trails are described as immutable on every pricing tier, audit evidence is captured on every change, and every action is "traceable back to its originating AI intent", so a reviewer can reconstruct not only what happened but which agent request caused it.

The MCP Server receives, validates and translates every request before anything touches infrastructure, applying RBAC, OAuth and audit logging automatically, so the agent's inbound intent and the resulting execution are captured on the same spine. Output, telemetry and result state are captured per task.

Insights adds measurement, with activity dashboards and SLA metrics on FlowAgent and workflow activity, performance and ROI. The audit trail ships on all five tiers, while Insights dashboards begin at Scale.

Sourceitential.com/cloud-platform/architecture, /pricing and /cloud-platform/itential-mcp-serverread 2026-09-08

Itential documents persistent orchestration state but no agent memory. Lifecycle Manager orchestrates "multi-day sequences with explicit state and full rollback", and stateful orchestration with lifecycle intelligence tracks infrastructure services with AI enhanced resource models that detect drift and automate changes from provisioning through retirement. Both are real persistence, but the state held is the orchestration's position in a running sequence and the resource model of the customer's infrastructure, the application's own data model: deleting it deletes the service record.

The vendor's own unit of work confirms the boundary. The pricing page defines an agent run as "one full instance of the agent executing its instructions, including all reasoning and tool usage, ending in a result", a session that starts and finishes, with nothing documented carrying across to the next. Itential also states it does not create copies of data, which cuts against an agent side store by design. No memory scope, stated lifetime, retention or deletion control over agent memory, or taught preference store carried between sessions is documented.

Sourceitential.com/pricing and /cloud-platform/architectureread 2026-09-08

Three customer environments are named in Itential's FAQ on deployment models, "on-premises, private cloud, or customer-controlled public cloud", followed by a plain commitment: "data and execution remain in your environment." The architecture page adds a self managed option for organizations that need complete control to satisfy regulatory or internal security requirements, alongside a fully managed SaaS offering.

The topologies are documented, not only the choice. Itential Validated Designs cover single server high availability, active/standby with full component redundancy tolerating a catastrophic failure, and blue/green multi region disaster recovery, with a developer architecture named separately as unsuitable for production.

The pricing matrix adds a 99.9% availability SLA and custom VPN connectivity, a dedicated private network path into regulated environments, both on the Enterprise tier. Itential names the regulatory driver itself: national governments and industry regulators placing controls on where customer data is physically located. All of this concerns where Itential itself runs, separate from the hybrid and multi cloud infrastructure it orchestrates.

Sourceitential.com/cloud-platform/architecture and /pricingread 2026-09-08

Adoptable units sit in a browsable catalog. The Itential Marketplace is an open source library of API integrations, "pre-built skills, agents, and workflows", with customers able to contribute their own. It is listed as unlimited on every one of the five pricing tiers, open source and never metered, and sits at itential.com/marketplace with a community forum alongside it and public GitHub and GitLab organizations behind it. The units are agents and workflows, not connectors alone.

A build path sits alongside it. FlowAgent Builder creates governed, purpose built agents from natural language descriptions with defined personas, scopes and toolsets; Itential Builder Skills and Spec-Driven Development generate automation from intent and convert a proven agent session into a deterministic workflow without a rebuild. Golden workflow patterns and Configuration Manager golden templates supply reusable starting points, and the ServiceNow App publishes governed automations into the customer's ServiceNow catalog.

Sourceitential.com/pricing, /marketplace and /itential-builder-skillsread 2026-09-08

Itential names its trigger surface as a capability: Event Triggers and Closed-Loop Automation connects FlowAI agents to observability and AIOps platforms for event driven orchestration with autonomous remediation and human in the loop governance, so a Selector, Forward Networks, Datadog or Splunk detection starts a workflow with no operator, and the loop closes back through the same governed path.

Four further entry points are documented.

ITSM tickets in ServiceNow and Jira resolve through integration, with the ServiceNow App embedding Itential actions as native Flow steps; CI/CD pipelines in GitHub and Jenkins trigger builds and deployments; the Operations Manager self service portal lets NOC, operations and service desk staff launch and monitor automation without developer access; and external LLMs invoke platform workflows as governed tools through the MCP Server.

Itential states the shape directly: infrastructure actions can be triggered by "humans, automation systems, or intelligent agents", through workflows published as microservices on an open API. Lifecycle Manager adds time extended sequences running over multiple days.

Sourceitential.com/cloud-platform/architecture and /pricingread 2026-09-08

Customer model choice gets its own row on the pricing matrix: the "LLM provider" row lists "Anthropic, OpenAI, Google, Azure, Bedrock, Ollama. Bring your own model, or use Itential's." The FAQ states the mechanism.

With a customer provided LLM, the customer configures a compatible provider on an account they control and pay for, naming Anthropic, OpenAI, Google, Mistral, Ollama, Azure OpenAI and AWS Bedrock.

With the Itential provided option, available only on Essentials and Compliance, Itential includes Anthropic as a turnkey part of the subscription, while Scale, Enterprise and Custom require a customer supplied model.

The FlowAI page adds the credential surface: provider API keys are managed through LLM provider profiles, write only once saved, masked in the UI and never returned through API or export. It also states support for any LLM, SLM or open weight model, including a custom one. Named providers, a self hosted open weight option through Ollama, customer held keys in a managed credential store, and a choice set out per tier put the decision with the customer.

Sourceitential.com/pricing and /cloud-platform/flowairead 2026-09-08

Extension surfaces run in both directions. Inbound, the Itential MCP Server exposes governed automation as tools "any LLM can discover and invoke" through the same policy enforced engine, sitting between AI systems and the platform to receive, validate and translate every request, and the FlowMCP Server is its enterprise form, centrally managing multiple MCP instances through virtual MCP servers with persona based access control and multiuser authentication. Outbound, the FlowMCP Gateway invokes external MCP tools.

Behind it is a documented API estate: a developer documentation site at docs.itential.com with release notes, and an open API through which workflows are published as microservices to northbound systems.

Every API already integrated with Itential is an agent tool with no new MCP server required, every self service catalog item is automatically registered as an MCP tool, and internal APIs and automation endpoints can be registered as tools.

Around it sit public GitHub and GitLab organizations, an Itential MCP product page, Builder Skills, native git connecting the customer's own repos, a community marketplace and an AWS Marketplace listing procurable against committed AWS spend.

Sourceitential.com/cloud-platform/itential-mcp-server, docs.itential.com and /cloud-platform/flowairead 2026-09-08

Workflows are tested before deployment, with an approval gate in front of production. Workflow Studio, its own line on the pricing matrix from the Scale tier, is a drag and drop builder "with versioning and testing" for designing and testing workflows in a visual low code canvas.

The vendor states the gate explicitly: "all agent-proposed automations pass through validation, testing, and approval processes before production deployment", and human review plus guardrails ensure only validated, compliant workflows execute in live environments.

Spec-Driven Development converts a proven agent session into a deterministic workflow, so a session that has been exercised becomes the artifact that ships.

What is tested is the workflow artifact, not the agent's reasoning, and no scored comparable result, holdout, benchmark or drift score is published: a customer sees pass or fail against their own logic rather than a graded evaluation. Pre and post validation with audit evidence and rollback on each change checks whether a change installed correctly, and Insights activity dashboards are a live operations metric.

Sourceitential.com/pricing and /cloud-platform/architectureread 2026-09-08

No hosted or local browser, desktop session, or remote or local computer control is documented. The agent reaches infrastructure programmatically: the Itential Gateway executes Python scripts, Ansible playbooks and OpenTofu plans against devices and cloud estates over API and CLI with secure multi vendor connectivity, and every other path, MCP, the northbound open API, ITSM and CI/CD, is programmatic. Terminal and shell access are not computer use, and normalizing CLI, API and AI interfaces into governed workflows is the opposite of operating a screen.

Nothing navigates, clicks, fills or advances anything through an interface built for a person. A device CLI session is a machine protocol, not a rendered surface, so the interface changes that break screen automation do not touch an SSH session to a router or a Terraform plan. For an infrastructure orchestration platform, that is the expected shape rather than a gap.

Sourceitential.com/cloud-platform/architecture and /pricingread 2026-09-08