Agent200 glossary: the terms, in plain words

A shared vocabulary makes everything else easier, from reading the docs to explaining your architecture to a teammate. These are the terms you will meet when building with Agent200, in plain words.

Agent200

Unified infrastructure for AI agent developers: models, connectors, managed OAuth, reusable workflows and paid services through one SDK and one API key.

Developer

You. You create the account and projects, choose the services and capabilities, and integrate the SDK into your application or agent.

End user

The person using the application you built. They do not need an Agent200 account. They authorize access to their own external accounts.

Project

A container for one application’s configuration in Agent200.

Environment

A separate configuration inside a project, such as development, staging or production, each with its own settings, integrations and credentials.

API key

The credential your application uses to call Agent200. It gives access to the capabilities permitted in its project or environment, and to every supported model and paid service, without separate provider keys.

SDK

The library you add to your application to make Agent200 requests.

Connector

Agent200’s integration with an external service. It handles authentication, endpoints, requests and token refresh for that service.

Premium connector

A paid connector managed by Agent200 that does more than relay an API. iGPT, which indexes email and answers with cited context in one request, is one.

Endpoint

A single operation a service offers, such as “search messages”. You choose which endpoints your application may use.

Tool

A capability the model can call. It has a name, a description and inputs. Behind it can be one endpoint, several, or a workflow.

Workflow

A named, reusable capability that combines endpoints from one or more services. Aggregated workflows combine results; sequential workflows pass one step’s output to the next.

Capability

Anything your agent can do through Agent200: a connector operation, a workflow, a model, a paid service.

run()

The request that executes a task. You name the user, model, instructions, input and allowed tools; Agent200 runs the tool-calling loop until the task is complete.

Tool-calling loop

The cycle inside a run(): the model asks for a tool, Agent200 executes it, the result goes back to the model, and so on until the model answers.

Managed OAuth

Agent200 runs the authorization process for supported connectors: authorization URLs, callbacks, token storage and refresh.

Connection

An end user’s authorized account on an external service, stored by Agent200 and associated with that user.

Scope

A permission requested from a provider during OAuth, such as read-only access to email. You choose the scopes; the user approves them; the provider enforces them.

Paid service

A third-party service that charges for usage, such as web search, reached with your Agent200 key and billed on your Agent200 statement.

Missing a term? Ask us through the contact page, and see the full picture on the Platform page.