> ## Documentation Index
> Fetch the complete documentation index at: https://docs.altrum.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Glossary

> What the words in Altrum mean.

| Term | Meaning |
| - | - |
| Access review | A snapshot of every key with when it was last used and what looks wrong, and a Keep or Revoke for each, kept as evidence. See [Access reviews](/access-reviews.md). |
| AI assets | The models and MCP servers your organization uses, governed under Govern in the sidebar: the [Model catalog](/model-catalog.md) and [MCP servers](/mcp-servers.md). |
| AI Gateway | The address your applications and coding agents send model requests to, so Altrum can govern each one. |
| AI system | Something your organization builds or buys that uses a model. |
| AI system owner | The member named as an AI system's owner. They can run its assessment and configure its guardrails whatever their role. |
| Alert | A notice that a rule held: a budget level reached, a spike in violations, an AI system waiting for approval. It clears by itself when the rule stops holding. See [Alerts](/alerts.md). |
| Always applied | A risk Altrum applies to every AI system, with controls Altrum enforces. |
| Apply | Make the saved guardrail draft the next live version. |
| Approved | An AI system an owner or Governance Lead approved, not yet live, or paused. |
| Approved models only | A switch in the Model catalog: when it is on, the AI Gateway serves approved models only, on every key. See [Model catalog](/model-catalog.md#serve-approved-models-only). |
| Assessment | A conversation with Altrum's assistant that finds the risks of an AI system. |
| Auditor | The role that reads the governance record (compliance reports, the audit trail, access reviews, the Model catalog, MCP servers and the members) and changes nothing. It sees no usage or spend, and keeps access until an owner or admin removes it. |
| Bill of materials | What an AI system used through the AI Gateway in the last 30 days: models, MCP tools and prompts. See [Discover](/ai-systems.md#the-ai-systems-page). |
| Cache savings | An estimate of what requests answered from the cache would have cost, each priced at the same model's other requests. See [Reporting](/reports.md#cost-and-usage). |
| Complete Assessment | Make the saved guardrail draft live and close the open assessment. |
| Compliance posture | Your organization's posture on [Home](/home.md): the share of the requirements for each AI system's risk level that are met, averaged over the AI systems not retired. |
| Compliance report | The evidence for one AI system or the organization over a period, on one page that prints and exports. See [Compliance reports](/compliance.md). |
| Control | A measure that answers one or more risks: preventive, detective, mitigating or administrative. |
| Control effectiveness | On [Home](/home.md), the share of the built-in attacks your organization's guardrails caught in their latest [red-team run](/guardrails/red-team.md). |
| Coverage | How much of your organization's AI Gateway traffic is registered (sent by an AI system) and guarded (checked by a guardrail). See [Discover](/discover.md#coverage). |
| Discover | Where traffic on keys with no AI system is listed, to register, attach or dismiss. See [Discover](/discover.md). |
| Draft | The guardrail settings you are working on, not enforced until applied. Also the first stage of an AI system, before it is approved. |
| EU AI Act category | Prohibited, High risk, Transparency obligations or Minimal, worked out from an AI system's risk profile. See [Risk rating](/ai-systems/risk-rating.md). |
| Go-live checks | What an AI system needs before it is approved: the risk profile, and at High and above a recent assessment, live guardrails and a second person. |
| Governance Lead | The role that approves AI systems, models and MCP servers and runs incidents and access reviews, without managing members, credentials or gateway settings. |
| Guardrail | A check on what people send to an AI system or what the model answers. |
| High risk after controls | On [Home](/home.md), the AI systems still rated High or Very high once their controls are counted. |
| Home | The first screen after you sign in: what needs you, the figures that matter, and the way to the detail behind each. See [Home](/home.md). |
| In review | An assessment that has been submitted and is open for review. Also an AI system waiting for approval. |
| Incident | A problem someone is handling, with an owner, a severity, a status and a timeline of what was done. See [Incidents](/incidents.md). |
| Initial rating | An AI system's rating when it was registered. It never changes. |
| Live | The stage in which an AI system takes traffic through the AI Gateway. See [Lifecycle](/ai-systems/lifecycle.md). |
| Live version | The guardrail version in force for an AI system. |
| Log only | A guardrail records what it found and lets the text through. |
| Mask | A guardrail replaces the sensitive part of the text. |
| Member | A person in your organization. Also the role for people who build and use AI: they register and run their own AI systems and use personal keys. See [Members and roles](/members-and-roles.md). |
| Model catalog | Every model your provider credentials can serve, with its price, use and status: Approved, Under review, Blocked or Not reviewed. See [Model catalog](/model-catalog.md). |
| Needs you | The part of [Home](/home.md) that lists the five things you can act on first. |
| Needs your action | A guardrail in an assessment's scope that is switched off or has nothing chosen yet. |
| Organization | Your company in Altrum. |
| Owner, Admin | The roles that run the organization in Altrum. An owner does everything, approves AI systems and receives the emails about the organization and its AI systems; an admin runs Altrum and the AI Gateway and approves nothing. |
| Personal key | A virtual key that belongs to one person. |
| Posture | How far an AI system meets what its rating asks of it, as a score. An AI system meets its requirements at 100%. See [Posture](/ai-systems/posture.md). |
| Project | A group of AI systems with an optional lead and members. Grouping only: every member still sees every AI system. See [Projects](/projects.md). |
| Projected this month | Spend so far this month divided by the days gone, times the days in the month. See [Reporting](/reports.md#cost-and-usage). |
| Provider credential | Your organization's own key for a model provider, stored in Altrum. |
| Rating | An AI system's current risk level, Low, Medium, High or Very high: the level a person chose, or else the one its risk profile gives. |
| Red-team run | The built-in attack set checked against live guardrails, showing what got through and the setting that would catch it. See [Red-team runs](/guardrails/red-team.md). |
| Reporting | What your AI traffic through the AI Gateway costs and how it is used, with guardrail violations and tool use. See [Reporting](/reports.md). |
| Residual rating | What is left of the rating once the live guardrails are counted: one level lower when every risk the latest assessment found is covered. |
| Retired | An AI system taken out of use: read only, its keys refused, until it is restored to Draft. |
| Risk | Something that can go wrong with an AI system. |
| Risk profile | Seven answers about an AI system that set its rating and EU AI Act category. |
| Roll back | Apply an earlier guardrail version again, as a new version. |
| Scope | The risks, controls and guardrails an assessment found. |
| Serious incident | Under the EU AI Act, an incident with a high risk AI system that leads to a death or serious harm to health, serious disruption of critical infrastructure, a breach of fundamental rights obligations, or serious damage to property or the environment. It is reported to the market surveillance authority within 15 days. See [Serious incidents](/incidents.md#serious-incidents). |
| Tool risk | What an MCP tool can do: Read, Write or Destructive, taken from what its server says, or set by a person. See [MCP servers](/mcp-servers.md#what-each-tool-can-do). |
| Version | A numbered, unchangeable snapshot of applied guardrail settings. |
| Violation rate | Requests a guardrail fired on, for every 1,000 requests the guardrails checked. Requests no guardrail checked are left out. See [Reporting](/reports.md#guardrail-violations). |
| Virtual key | The key an application or a person sends to the AI Gateway. |


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.