Terminology

Plain definitions for the terms you'll meet in these docs and in the Argus app. If a word on another page links here, this is where it lands.

Agent

An AI assistant you direct to get work done, like Claude Code, Codex or Claude Cowork. You give it a goal in plain language and it carries out the steps, using tools to read files, run commands or call other systems.

Argus Hub

A self-hosted server that collects usage from many people's Argus and shows an org-wide view. It's how a team or ops leader sees how agents are used across a group, rather than just their own use.

Cost

The estimated dollar cost of your agent use, worked out from how many tokens you used and each model's price. It's an estimate for understanding trends, not a bill.

Dashboard

The org-wide view an Argus Hub presents, pulling together usage from everyone whose Argus syncs to it. It's how an ops leader sees agent use across a team, rather than one person's own usage in the Argus app.

Friction

Signs that a session didn't go smoothly: interruptions, tool actions you declined, context that had to be compacted and slow turns. Argus surfaces these on the Health view so you can see where your agents get stuck. Friction is measured for Claude sessions only.

Index

The local store Argus builds from your sessions. Indexing goes through each session once, pulls out the useful details (usage, cost, tools, skills and the like) and saves them so the app is fast and your sessions are ready to explore. The index lives on your own computer. Nothing is uploaded unless you choose to sync.

Interaction

One back-and-forth with an agent: a prompt you send and the reply it works out, including any tools it uses along the way. Argus counts interactions per session and per task as a measure of how much work each one took.

MCP server

A connector that gives an agent extra abilities, like access to a database, a ticketing system or a web service. MCP (Model Context Protocol) is the shared standard agents use to talk to these connectors.

Model

The specific AI behind an agent, such as Claude Opus, Claude Sonnet or a GPT model. Different models have different speed, capability and price, so your usage and cost depend on which ones you use.

Model gateway

An OpenAI-compatible endpoint that sits between Argus and a model provider. The gateway holds the provider's API key, and Argus reaches it with a separate key, so the provider key can stay off your computer or on a shared server your team runs. See Model Gateway.

Plugin

A bundle that adds skills, tools or connectors to an agent in one package. Argus can show which plugins you have enabled and which ones you aren't actually using.

Project

A body of work an agent associates with a folder on your computer, usually one codebase or working directory. Argus groups your usage by project so you can see where your agent time goes.

Session

One continuous working conversation with an agent, from the first thing you ask to when you stop. A session can include many back-and-forth turns and the tools the agent used along the way.

Skill

A reusable set of instructions an agent can pull in to do a particular kind of task the same way each time, like drafting a release note or formatting a report.

Source

Which agent a piece of usage came from. Argus indexes Claude Cowork, Claude Chat, Claude Code, ChatGPT Work, Codex and Gemini CLI. Claude Chat usage is estimated rather than metered, and it stays on your machine: it isn't uploaded to an Argus Hub.

Sync

Uploading a snapshot of your usage to an Argus Hub so it can be combined with other people's. Nothing is uploaded unless you choose to sync; indexing and the local app stay on your machine.

Task

A single thing you set out to do in a session, like researching an account or drafting a post. When you turn on task interpretation, Argus groups a session's back-and-forth into tasks and judges how each one went.

Token

The unit of text an agent reads and writes. Usage and cost are measured in tokens, so a longer conversation or a bigger file means more tokens.

Tool

A single action an agent can take, such as reading a file, running a command, searching the web or calling an API. Argus shows which tools your agents use most.