Core concepts
The words you will see in Surface, explained once, in plain English.
Surface uses a small set of words over and over. This page defines each one. Other pages link here instead of explaining the terms again.
Endpoint
A computer with the scanner installed. Usually an employee laptop, running macOS, Windows, or Linux. The Endpoints page lists all of them.
An endpoint is Active when it reported recently and Stale when it has stopped reporting.
Asset
One AI tool the scanner found. Claude Code on a developer’s laptop is an asset. So is a ChatGPT browser add-on, a model file, or a connector that gives an agent access to GitHub.
Every asset has an asset type:
| Type | In plain words | Examples |
|---|---|---|
| Agent | An AI program that acts on its own | Claude Code, Cursor, Codex CLI, ChatGPT desktop, Ollama |
| MCP Server | A connector that gives an agent access to something: files, GitHub, a database. MCP is the Model Context Protocol | Filesystem, GitHub, Slack connectors |
| Skill | Instructions that teach an agent a task, usually a folder of text files | Skill folders in agent settings |
| Browser Extension | An AI add-on inside Chrome or Firefox | AI writing assistants |
| IDE Extension | An AI add-on inside a code editor | GitHub Copilot, Codeium, Continue |
| Model | The AI model itself, stored as a file | Models from Hugging Face or Ollama |
| Memory | Where an agent stores what it remembers | Local vector databases |
| Tool | A function an agent is allowed to call | Tool definitions in agent settings |
Instance
One copy of an asset on one laptop. If Cursor is on 40 laptops, the console shows one asset with 40 instances.
The Inventory page calls this count Instances. The Remediation page calls it Endpoints. Both mean the number of laptops that have the tool.
Inventory
The full list of assets found anywhere in your company. Each asset appears once, however many laptops have it.
Connection
A link between two assets. Claude Code using three MCP servers has three connections. The Asset Graph draws them.
Policy
One rule that describes a risky situation, such as “an MCP server has a password saved in plain text”. Surface ships with about 200 policies. SuperAlign writes and updates them. You can read them on the Policies page but not change them.
Finding
An asset breaking a policy. The console calls findings Failed Policies. They are the only thing that decides an asset’s risk score.
Risk score and risk level
Every asset gets a score from 0 to 10 and one of four levels:
| Level | Score | Meaning |
|---|---|---|
| Critical | 8.1 to 10 | Act now |
| High | 6.1 to 8 | Review soon |
| Medium | 4.1 to 6 | Keep an eye on it |
| Low | 0 to 4 | Fine under normal rules |
An asset with no findings shows a dash instead of a score. See How risk is calculated.
Governance status
Ungoverned means nobody has decided about the asset yet. Governed means someone chose Allow, Warn, or Block. New assets start Ungoverned and wait in the Open list on the Remediation page.
Allow, Warn, Block
The three decisions you can make about an asset:
| Decision | On the laptop |
|---|---|
| Allow | Nothing changes. The tool is approved |
| Warn | The user sees a message that the tool is not approved. It stays installed |
| Block | The scanner removes the tool. If it cannot, it closes the tool whenever it opens |
See Governance.
Session
One conversation between a person and an AI tool. Surface records sessions only if you turn that on. The message text is stored separately and opening it is logged.
Extension
An optional feature you switch on. Prompt collectors run on laptops and record sessions from one AI tool. Cloud extensions send session data to storage you own, such as S3 or Splunk. See Extensions.
Integration
A connection to another company service, such as Google Workspace or Okta, to keep your user list in sync.
Enrollment token
A long secret code that lets a laptop join your organization without anyone signing in. You create it in the console and hand it to your MDM.
Organization
Your company’s private space in Surface. Nothing in it is visible to any other company.
Role
What a person may do in the console: Owner, Admin, Viewer, or Member. See Users and roles.
Audit log
The record of who did what and when in the console. It cannot be edited.