.AICore concepts
Definitions of the objects APIANT.ai is built from, including automations, triggers, steps, actions, runs, connections, apps, connectors, assemblies and subroutines.
An automation is a trigger followed by an ordered list of steps, stored as one document on your account. Every other object on this page is either part of an automation, something an automation uses, or the record of an automation running.
Automation
An automation has a name, a folder, a trigger (some have more than one), and the steps that run after it. It is either on or off. Its trigger fires only while it is on, though a test run works either way. A new automation is saved turned off until you, the assistant or Claude turn it on.
Saved changes are recorded as versions, so you can see what changed between them. You build and change automations by asking the assistant or Claude. The diagram in the app shows the automation read-only, and you can leave a comment on any step for Claude to apply.
Read more: Automations, Drafts and versions, Review a diagram and send comments to Claude.
Trigger
A trigger decides when an automation runs and supplies the data the first step receives.
| Trigger | Runs when |
|---|---|
| Polling | On a schedule, APIANT asks an app for new or updated records, and the run processes each record it has not seen before |
| Webhook | An app or program sends an HTTP request to the automation's webhook URL |
| Schedule | A time or recurring schedule comes around |
| Web service | A caller requests the automation's web service URL and waits for the response |
| Form | Someone submits the automation's hosted form |
| Chat | A visitor sends a message in the automation's chat widget |
| Tool input | An AI agent or MCP client calls the automation as a tool |
| Manual | It is started directly, with no event source |
On a step's detail dialog, a trigger from an app carries an INSTANT or POLLING badge. Hovering the badge names how the trigger's events reach APIANT.
Read more: Triggers.
Step and action
A step is one unit of work after the trigger. Most steps are actions: an operation on an app, such as creating a HubSpot contact or posting a Slack message, or a built-in operation such as formatting a date. Other steps control the flow of the automation: conditions, loops, parallel branches, waits, and calls to other automations.
A step reads data from the trigger and from earlier steps through field mappings, and its output is available to every step after it.
Read more: Actions and built-in apps, Field mapping, Conditions, loops and parallel branches.
Run
A run is one execution of an automation. It records the trigger data, each step's input and output, and whether it succeeded, failed or is waiting. Runs are numbered per automation (Execution #1, #2 and so on) and listed in Execution history. A run that waits on a form, an approval or a delay is paused and resumes later.
Automations that run use the tasks included in your plan.
Read more: Execution history, Retries, replays and restarts, Plans, usage and billing.
App, connector and connection
An app is a service in the APIANT catalog, such as Shopify or Slack. Each app has a connector, which defines how APIANT authenticates to it: OAuth, an API key, or another scheme the vendor requires.
A connection is your authorized account on an app. You create one on the Connections page, or from the Connect button the assistant shows. A step that uses an app runs with one of your connections for that app.
Read more: Apps and connectors, Connecting accounts.
Assembly
An assembly is the reusable unit behind a trigger or action: the API requests and logic that turn "create a contact" into calls to the vendor's API, with typed inputs, settings and outputs. The catalog's triggers and actions are assemblies. You do not need to work with assemblies to build automations; when an operation is missing and your plan allows app authoring, Claude builds or edits them in Claude Code.
Read more: Build a connector, action or trigger.
Subroutine
A subroutine is an automation that other automations call, optionally with declared inputs and outputs. Error handlers are subroutines too: a step or an automation can name one to run when something fails.
Read more: Subroutines and calling automations, Error handling and retries.
The assistant and Claude
The assistant is the APIANT assistant in the app, under Assistant. Claude is Anthropic's model working in Claude Code, connected to your account through the APIANT MCP server. Both build and change automations by calling APIANT tools on your account; they differ in where they run, what they can do and what they cost.
Read more: The APIANT assistant, APIANT over MCP.