Key Concepts ​

This page is a reference you can return to at any point. Every other section of the docs builds on the terms defined here — knowing them up front will make the rest of the documentation easier to follow.

Agent ​

An agent is a lightweight binary you install on a Windows, Linux, or macOS host you own. Once running, it connects to InfraScout over mutually authenticated TLS, registers itself, and sends periodic heartbeats so the platform knows it is online. When an assessment session is active, the agent receives commands and executes them on the host, returning output to InfraScout. You install and manage agents on your own infrastructure; InfraScout never initiates a direct connection to your hosts.

For installation and configuration details, see Agents.

Enrollment Token ​

An enrollment token is a one-time credential you generate in the InfraScout dashboard. You supply it to the agent during initial setup so InfraScout can verify the agent's identity and issue it a signed TLS certificate. Once enrollment is complete, the token cannot be reused. If an agent is removed and re-enrolled, a new token is required.

Agent Capability ​

A capability is an agent's statement that it can honor a specific kind of command — running a shell, querying WMI, reading the systemd journal, capturing an ETW trace, hosting an MCP server. Assessment tools that require a capability only target agents that advertise it, and an agent that lacks it rejects the call.

Capabilities are detected on the host, not assumed from its operating system: most are probe-backed, so a Windows host advertises wmi only when the WMI subsystem actually answered the agent's probe, and a Linux host advertises journald only when the system journal is genuinely readable. Thirteen capabilities exist, spanning shell and PowerShell execution, Event Log, WMI, and LDAP queries, log-file and journal reads, performance data, ETW tracing including boot-time capture, long-lived daemon starts, and MCP hosting.

The full table — each capability, its platforms, what has to be true on the host for it to appear, and which tools require it — is on Agent Capabilities.

Agent Group ​

An agent group is a logical collection of agents used to scope assessments and manage access. Groups can be static, where you assign agents by hand, or dynamic, where InfraScout assigns agents automatically based on rules such as operating system, hostname pattern, or reported capabilities. A single agent can belong to more than one group.

For details on creating and managing groups, see Agent Groups.

Session ​

A session is a scoped assessment run against a specific agent. It tracks everything that happens during the assessment: the commands executed, their timing, their output, and the insights produced. A session is opened by InfraScout via the session_start MCP tool and finalized with session_submit. Sessions give every assessment a clear boundary, making findings reproducible and auditable.

For more on running assessments, see Running Assessments.

Session Group ​

A session group bundles the per-agent sessions that belong to a single piece of work — one chat conversation, one workflow run, or one direct MCP connection. An assessment that touches five hosts opens five sessions, and the group is how you see them as the one engagement they are. Groups carry a kind, a status (active, completed, interrupted, or failed), and a summary, so the Session Groups view reads as a list of work rather than a flat list of sessions.

Execution ​

An execution is a single command run within a session. InfraScout records the command text, the output it returned, how long it took, and whether it succeeded or failed. Executions are the raw evidence that backs up insights — when InfraScout or a playbook surfaces a finding, the supporting executions are always available for review.

Assessment PowerShell can also run in a validated read-only mode. When the AI uses the read-only PowerShell tool, InfraScout checks the script against an admin-managed command policy before dispatch and blocks anything that appears to change host state, so investigative work carries a guarantee that nothing it runs can modify the target. See Read-Only PowerShell for how it works.

Insight ​

An insight is a structured finding that InfraScout or an assessment playbook saves during a session. Every insight carries a severity level (critical, high, medium, low, or info), a category (security, identity, networking, compliance, performance, configuration, availability, or licensing), and a status that tracks its remediation lifecycle. Status progresses from open through acknowledged and in_progress to either resolved or dismissed.

For guidance on working with insights, see Insights.

Playbook ​

A playbook is a markdown document that guides an AI through a structured assessment. InfraScout exposes playbooks as MCP resources, so the connected AI client can retrieve and follow them during a session. Built-in playbooks cover common scenarios like Entra ID hygiene and endpoint compliance; you can also write your own to match your organization's specific requirements.

For authoring guidance, see Playbooks.

MCP Tool ​

An MCP tool is a function InfraScout exposes over the Model Context Protocol. AI clients call these tools to list agents, start and manage sessions, execute commands, save insights, and query your Microsoft cloud environment. InfraScout registers 207 tools in total, organized by prefix: agent_* tools manage agents and groups, session_* tools control assessment sessions, insight_* tools create and update findings, and mscloud_* tools interact with Entra ID, Azure, Microsoft 365, and Microsoft Defender.

Every tool declares how it behaves — whether it only reads, whether a change it makes is reversible, and whether it reaches outside InfraScout's own records. Those declarations are what confine standard users to read-only tools. See Understanding MCP for the protocol itself and Tool Annotations and Role Scoping for how the classification becomes an enforced boundary.

Tool Group ​

A tool group is a named bundle of tools a user grants the AI in one decision — "Entra ID Identity" rather than dozens of per-tool checkboxes. Built-in groups ship with InfraScout, each registered external MCP server appears as its own group, and custom groups are bundles you compose yourself from any mix of the two. Every group is scoped to Entra security groups, so which groups a person can reach is an administrative decision, enforced on every path a caller connects through.

For composing and scoping groups, see Tool Groups (Admin).

Task Agent ​

A Task Agent is an isolated run the AI delegates a self-contained piece of work to. It receives a brief and a scoped set of tools, works in its own fresh context, and reports back only a distilled result — a summary, findings, and the references behind them — so the raw material it churned through never crowds the main conversation. Delegated runs are read-only by default, several can run in parallel within one turn, and each keeps its full transcript for audit.

See Chat for how plans and delegation appear in a conversation.

Agentic Workflow ​

An agentic workflow is an assessment that runs without a person in the loop. You define it once — an instruction, the tool groups it may use, the service-account identity it runs as — and InfraScout fires it on demand, from a webhook, or on a schedule. The run executes headless, files its insights, and lands in a review queue. Because no human approves each step, the boundaries set on the workflow are the boundaries the run lives inside.

For configuration, see Agentic Workflows.

Persistent Memory ​

Persistent memory is how the AI keeps context across conversations. Instead of starting every chat from scratch, it reads and writes entries in a tenant-scoped store — what your environment looks like, what past assessments established, where work left off — and a background rollup pass keeps idle threads summarized. Memory is always available to the AI, and administrators can inspect every entry.

For the inspector and rollup behavior, see Memory (Admin).

Entra Connection ​

An Entra connection links your Microsoft Entra ID (formerly Azure AD) tenant to InfraScout. Once the connection is established, InfraScout synchronizes your users, groups, and group membership. This synchronization is what unlocks the full Microsoft Cloud tool set — without it, mscloud_* tools that query your tenant are unavailable.

For setup instructions, see Entra Connection.

Visibility ​

Visibility is the access control layer that determines which resources — agents, agent groups, tool groups, playbooks, and AI provider connections — a given user can see and interact with. You scope a resource to one or more Entra groups, and only members of those groups can see it; admins always see everything.

Visibility is closed by default: a resource with no groups assigned is visible to admins only. Mark it Everyone when it should be visible tenant-wide. Agents inherit visibility from the agent groups they belong to, so scoping a group covers its members — unless you turn on block inheritance on an individual agent to keep a sensitive host hidden even inside a broadly visible group.

For the full model and the administration surface, see Groups (Admin).