Home / Glossary / ChatGPT Actions
AI · Updated May 10, 2026

ChatGPT Actions

ChatGPT Actions are OpenAI's mechanism for letting ChatGPT — including Custom GPTs — call external APIs to fetch data or perform tasks during a conversation.

Also called: GPT Actions · OpenAI Actions

What it is

ChatGPT Actions extend a conversation by giving ChatGPT (or a Custom GPT) the ability to make HTTP calls to external services on the user’s behalf. Each Action is configured via an OpenAPI 3.x specification that declares the endpoints, their parameters, the authentication scheme, and the expected responses. When a user’s question requires fresh data or a real-world side effect, the model picks the appropriate endpoint, makes the call, and folds the response into its answer.

Actions replaced the older “ChatGPT Plugins” surface (deprecated in 2024) and are now the canonical way to extend ChatGPT inside the OpenAI ecosystem.

How they work

An Action is defined by three things: an OpenAPI spec describing the API, an authentication configuration (API key, OAuth, or none), and a privacy policy URL. Once installed inside a Custom GPT or available to ChatGPT through Enterprise tooling, the model can read the spec, decide when an endpoint is relevant, and call it. The result is returned as structured data the model summarizes.

Actions are scoped to a single Custom GPT or workspace. Each user authenticates against the underlying service independently — there’s no shared session.

Where it shows up

Most enterprise integrations into ChatGPT today land as Actions: CRM lookups (“what’s the latest opportunity for Acme Corp?”), document search (“find the SOW for project X”), data-warehouse queries, ticket creation, calendar reads, code-search across internal repos. Any system with a documented API can be wrapped.

How Aquifer fits

Aquifer ships a ChatGPT Action that surfaces a customer’s connected data — Procore projects, Sage 300 CRE financials, ArcGIS spatial records, SharePoint document archives — through a single OData-style query surface. Configure once at the Aquifer level; the same surface backs both the Action (for ChatGPT) and the MCP server (for Claude, Cursor, and other MCP-compatible tools). Same data, two doorways, all the major AI tools covered.

See Your AI of Choice, Connected to Your Operational Data for the architecture and a customer example.

Want to see this in your stack?