Concepts
This page defines the terms the Faustine portal and these docs use, and links to the page that covers each one. For one-line definitions in alphabetical order, see the Glossary.
How the terms fit together
Inside a workspace, a Web Agent is built from a web app's code and an API is built from a service's code or spec. The analysis of a web app produces actions, grouped in modules. The analysis of an API produces operations, and you can add recipes on top of them.
A gateway serves the operations of one or more APIs to AI clients. A connection tells Faustine where the API runs, and a credential tells the gateway who is calling. Web Agents, APIs and gateways all change through versions and are served through environments. All the AI work runs on the workspace's model provider.
Workspace
A workspace is your company's account in Faustine. It holds your Web Agents, APIs and gateways, the people on your team, the billing and one model provider. A person can belong to more than one workspace. See Work in more than one workspace and Invite people to your workspace.
Web Agents
| Term | What it is |
|---|---|
| Web Agent | An agent inside one web app, built from that app's code. It has its own analysis, configuration, versions and environments. The portal's wizards and buttons call it "assistant". See Web Agents. |
| Action | One thing the agent can do in the app, such as deals.create. An action is generated code that runs in the page, in the user's session. You can turn an action off, edit it, or make the agent ask the user before it runs. See Actions. |
| Module | A group of actions that belong to one feature of the app, such as deals or invoices. You choose which modules to analyze. See Choose what to analyze. |
The user meets the agent in the widget, the chat window that the script tag or the npm package adds to your app. See Installation.
APIs
| Term | What it is |
|---|---|
| API | What Faustine knows about one of your services: its operations and their schemas, and how the service authenticates. It is the contract, not the running service. Every API has a short namespace, such as crm, that is set when the API is created and does not change. See APIs. |
| Operation | One call an AI client can make, such as crm.deals.create. An operation is a read or a write, and a write can require confirmation. See Operations. |
| Recipe | A composite operation: several calls to the API in one step, with its own input schema, written as a short script. See Recipes. |
| Connection | Where a service runs (its base URL) and the values of the credential Faustine uses to reach it. An API has one connection for each place the service runs, such as staging and production. See Connections. |
An API is served through a gateway.
Agent Gateway
| Term | What it is |
|---|---|
| Gateway | The address AI clients connect to. A gateway serves one or more APIs at one URL, over MCP and REST. "Agent Gateway" is the product, and a gateway is one that you create. See Agent Gateway. |
| Credential | Who may call one environment of a gateway. A credential has a scope, a behavior for writes and for operations that require confirmation, and limits. There are two kinds. See Scope a key or user. |
| API key | A credential for automations and code. The client sends it as a bearer token. Keys start with fik_. See Create an API key. |
| Gateway user | A credential for a person, who signs in from their AI client. See Create a gateway user. |
The Agent Gateway section has its own Concepts page with the order in which a gateway is set up.
Versions and environments
A version is a published, numbered snapshot that does not change. Changes you make go into the draft, and publishing turns the draft into the next version. Web Agents and APIs both have versions.
An environment is a named place where a version is served, such as Production or Staging. The term means slightly different things in the two products:
| Web Agent environment | Gateway environment | |
|---|---|---|
| Serves | One version of the agent | The APIs you add, each at a version |
| Has its own | Variable values and allowed domains | URL, exposed operations, connections, policies, API keys, gateway users, docs and logs |
| Chosen by | The env value your app passes when it loads the widget |
The URL the client connects to |
| Covered in | Versions, environments and deployments | Environments |
For API versions, see Versions and updates.
Model provider
The model provider is the AI service your workspace runs on: Anthropic, Amazon Bedrock, Azure AI Foundry, LiteLLM or Vercel AI Gateway. The analysis, the Advanced editor and the chat all use it, and it bills you for the usage. A workspace has one, set in Settings > Model Provider. See Model providers.
The model provider is not the same setting as an agent's chat model. The chat model is the model one Web Agent answers with, and each agent has its own. See Choose the chat model.
Next steps
- Quickstart: add an agent to your web app: see a Web Agent, its modules and its actions on your own app.
- Quickstart: let AI clients operate your API: see an API, a gateway and a credential working together.
- Glossary: every term on one page.
Last updated November 13, 2026