# Glossary

> Definitions for Endue terminology — agent, run, skill, connector, memory, output, routine, and the rest of the vocabulary you will meet in the product.

This glossary defines every term Endue uses. Each entry links to the page where the concept is covered properly. A term that appears in the product but not on this page is a documentation bug — please tell us.

## A

### Activity

The home screen of the web app: every [conversation](#conversation) across every [agent](#agent), grouped by whether it is waiting on you, running, or done. The place to look when more than one thing is in flight.

Learn more: [Activity](/docs/work/activity/)

### Agent

A durable identity that does work on your behalf, with its own [handle](#handle), profile, [character](#character), and [system prompt](#system-prompt). Unlike a chat window you open and discard, an agent persists: it keeps its abilities, its [memory](#memory), and its history. An agent can only reach what you have [bound](#binding) to it.

Learn more: [Core concepts](/docs/get-started/core-concepts/#agent)

### Agent Builder

The part of the web app where you create and tune agents — identity, system prompt, model, and the [skills](#skill) and [connections](#connection) bound to them.

The name is Agent Builder. "Studio" is not a product name and does not appear in the interface.

Learn more: [Agent Builder](/docs/build/agent-builder/)

### Approval

A pause in a [run](#run) where Endue asks you to confirm before the agent takes an action you cannot take back. Two categories always trigger it: **outbound** actions that leave your account and reach another person, and **destructive** actions that delete or overwrite data.

You see the exact action with the exact arguments before deciding. This gate is not configurable. In a [routine](#routine) or any other run nobody is watching, the action is refused rather than left waiting.

Learn more: [Approvals](/docs/work/approvals/) · [Core concepts](/docs/get-started/core-concepts/#approval)

### Automatic selection

A setting that has a decision model rate how hard each incoming message is, then answers with the model the agent assigns to that level. In Observe mode it only records the rating and leaves the model unchanged.

Learn more: [Choosing a model](/docs/build/models/#picking-a-model-for-each-request)

## B

### Binding

The link between an [agent](#agent) and a capability it is allowed to use. [Connections](#connection) and [skills](#skill) are bound to specific agents — holding a connection in your account does not grant it to every agent you own.

Binding is what keeps an agent's reach predictable. An agent with three tools behaves more consistently than one with twenty.

Learn more: [Connectors overview](/docs/connect/overview/#connection-versus-binding)

### BYOK

"Bring your own key." Using your own model-provider credentials for an agent's model calls instead of Endue's.

Learn more: [Bring your own key](/docs/account/bring-your-own-key/)

## C

### Channel

A pairing that lets people reach an agent from outside Endue — a Discord server or a Slack workspace. The opposite direction from a [connector](#connector), which lets an agent reach a service.

Learn more: [Channels](/docs/automate/channels/)

### Character

The disposition that shapes an agent's tone and style. Separate from the [system prompt](#system-prompt), which defines the job rather than the manner.

### Connection

An authorized link between your Endue account and an external service — your Gmail account, your Slack workspace, your Notion. A connection is tied to one account at the provider and can be revoked from either side at any time.

Connecting a service is not the same as granting it: a connection must be [bound](#binding) to an agent before that agent can use it.

Learn more: [Connectors overview](/docs/connect/overview/)

### Connector

The integration that makes a service connectable. Gmail is a connector; your authorized Gmail account is a [connection](#connection). Endue ships connectors for more than thirty services.

Learn more: [Available connectors](/docs/connect/available-connectors/)

### Conversation

A continuous, persistent thread with an agent. It holds the message history the agent reads for context and survives closing the app. Conversations can be grouped into a [project](#project) when several of them concern the same body of work.

Learn more: [Conversations](/docs/work/conversations/)

### Credit

The unit the [usage dashboard](/docs/account/usage/) reports weekly model spend in. The dashboard states the ratio it uses.

Learn more: [Usage dashboard](/docs/account/usage/#overview)

## D

### Device

A computer or browser an agent works on, or an app you use endue on. There are four device products: [endue Computer](#endue-computer), endue for Chrome, endue for Android, and endue Desktop. Every device has one of two subjects:

- **User**: a device you signed in to or registered with your account.
- **Agent**: a device an agent launched on endue cloud to make its own connection.

Devices are listed in the **Resources** panel, opened from the left sidebar.

Learn more: [Devices](/docs/build/devices/)

### Direct connection

A link between your agents and a server you run, either a remote MCP server or an HTTP API, set up with its endpoint and headers. Every call asks for [approval](#approval) unless you turn on Always allow for it.

Learn more: [Direct connections](/docs/connect/direct-connections/)

### Dream

A [trait](#trait) that tidies an agent's [memory](#memory) while it is idle. It merges overlapping memories, corrects stale facts, closes finished commitments, and archives memories that are no longer useful. It also gives each memory a one-line summary and search terms. It runs as a [routine](#routine) at the bedtime you set.

Learn more: [Traits](/docs/build/traits/#when-dream-sleeps)

## E

### endue Computer

The computer an agent keeps files on and runs programs in. The agent starts one on endue cloud the first time it needs it, or you give it your own Linux server or VM instead. Its file area is the agent's [workspace](#workspace).

Learn more: [Devices](/docs/build/devices/) · [Connect your own server or VM](/docs/build/own-server/)

## H

### Handle

The short name you use to address an agent. Generated automatically from its name when the agent is created, and stable afterwards.

## L

### LLM API

Calling a model directly from your own code with an Endue API key, in the OpenAI or Anthropic request format, paid from your [credits](#credit). No agent is involved — for an agent with its prompt and tools, use the Agent API.

Learn more: [LLM API](/docs/build/llm-api/)

## M

### Memory

What an agent retains across [conversations](#conversation) — facts about you, your preferences, how you like work done. The agent decides what is worth keeping and recalls it when relevant, so memory is selected rather than dumped into every [run](#run).

Memory is inspectable and editable: you can see what an agent stored and delete anything that should not have been kept.

Learn more: [Memory](/docs/capabilities/memory/)

## N

### Notification

A message telling you something finished while you were elsewhere: a [run](#run) replied, a [routine](#routine) completed, failed, or was skipped. Notifications can also arrive as browser notifications when the app is closed. They tell you what happened; you act on it in the conversation they point to.

Learn more: [Notifications](/docs/work/inbox/)

## O

### Output

Something an agent produced that is worth keeping, stored on its own page rather than buried in a [conversation](#conversation). An output is either a file (a document, report, diagram, or dataset) or an app (several files that work together). Outputs are private to you by default and can be shared with a link that grants access to that one output and nothing else.

Learn more: [Outputs](/docs/capabilities/outputs/)

## P

### Project

A grouping of [conversations](#conversation), [outputs](#output), and [routines](#routine) that belong to the same body of work. An agent working inside a project sees the project's context, which is the cheapest way to stop re-explaining background in every conversation.

Learn more: [Projects](/docs/capabilities/projects/)

## R

### Reasoning effort

How much thinking a model does before it answers. Higher effort costs more tokens and takes longer, and pays off on multi-step work where an early wrong turn wastes the rest of the [run](#run). Available on models that support it.

Learn more: [Choosing a model](/docs/build/models/#reasoning-effort)

### Resources

The [skills](#skill), [connectors](#connector), and [devices](#device) an agent can use. The **Resources** panel in the left sidebar holds the ones on your account. The Resources section of [Agent Builder](#agent-builder) shows what is [bound](#binding) to one agent, together with its built-in tools.

Learn more: [Agent Builder](/docs/build/agent-builder/)

### Revision

A saved version of an agent's [system prompt](#system-prompt). Prompts are versioned so a change can be reviewed and rolled back — which is what makes prompt iteration safe rather than a gamble.

Learn more: [System prompt and revisions](/docs/build/system-prompt/#revisions-and-rollback)

### Routine

A task an agent performs on a schedule, running whether or not you have Endue open. What a routine produces waits for you in your workspace. If it needs a decision it asks, and an [approval](#approval) it cannot get is refused rather than left hanging.

Learn more: [Routines](/docs/automate/routines/)

### Run

One turn of an agent working: read the context, decide, call a [tool](#tool), read the result, repeat until done or until it needs you. Runs execute on Endue's servers, so closing the app does not stop the work.

Learn more: [Runs](/docs/work/runs/) · [How Endue works](/docs/get-started/how-it-works/#the-loop)

## S

### Skill

A packaged, repeatable job — a procedure, a checklist, a set of commands — that can be [bound](#binding) to an agent so you do not have to re-explain it. An agent can also request a skill it has not been given, pausing mid-[run](#run) for you to grant or decline.

Learn more: [Skills](/docs/capabilities/skills/)

### Slash command

A name typed after `/` in the composer. Built-in commands such as `/usage` are app features and start no [run](#run); the rest come from the [skills](#skill) bound to the agent and tell it which [tool](#tool) to call first.

Learn more: [Slash commands](/docs/work/slash-commands/)

### Space

A shared workspace where a team uses the same agents and sees the work they produce.

### Steering

Sending a correction to an agent while a [run](#run) is still in progress. The agent picks it up on its next step, so a correction costs one step rather than the whole run.

Learn more: [Steering a run](/docs/work/steering/)

### Surface

One of the places you can use Endue: the web app, the desktop app, mobile, or the command line. All surfaces reach the same agents, conversations, and history — you can start on one and continue on another.

Learn more: [Surfaces](/docs/get-started/surfaces/)

### System prompt

The standing instructions an agent reads before every conversation: what it does, what it must never do, and what a good result looks like. Kept under [revision](#revision) history.

Learn more: [System prompt and revisions](/docs/build/system-prompt/)

## T

### Third-party skill

A [skill](#skill) imported from a public GitHub or GitLab repository. Only the text of its `SKILL.md` is imported, and the skill is private to your account.

Learn more: [Third-party skills](/docs/capabilities/third-party-skills/)

### Tool

A single capability an agent can invoke during a [run](#run) — read a file, search email, create a calendar event, write an [output](#output), recall a [memory](#memory). Tools are what make an agent able to act rather than only advise.

Some tools are built in and always present. Others appear only when a [connection](#connection) or [skill](#skill) is bound to the agent.

Learn more: [Built-in tools](/docs/capabilities/built-in-tools/) · [Core concepts](/docs/get-started/core-concepts/#tool)

### Tool call

One invocation of a tool within a run, visible to you as it happens, along with what it returned.

### Trait

A switch that turns one behavior of an agent on or off. The traits are Personality, Self-evolving, and [Dream](#dream). The setting stays where it already lives, and the trait decides whether it is used.

Learn more: [Traits](/docs/build/traits/)

## U

### Usage dashboard

The record of what your agents have spent — tokens over time, broken down by model and by [agent](#agent) — together with how much of your plan allowance is left.

Learn more: [Usage dashboard](/docs/account/usage/)

## W

### Workspace

An agent's own file area on its [endue Computer](#endue-computer): the folder it reads, writes, and runs programs in. The word means the folder, not the machine.

Learn more: [Workspace](/docs/build/workspace/)

---

## Retired terms

These appear in older material. Use the current name.

| Old term | Now called | Notes |
| --- | --- | --- |
| Studio | [Agent Builder](#agent-builder) | Never a product name; do not reintroduce it |
| Abilities, Ability | [Resources](#resources) | The old name of the same panel and Agent Builder section |
| Computer, workspace (meaning the machine) | [endue Computer](#endue-computer) | "Workspace" now means only the agent's file area |
| Artifact | [Output](#output) | An output that is a single file. Apps are outputs too |
| Character build, quest, party | — | The RPG framing is retired from product copy |