Reference
AI Agent Glossary: Dots, Agents and Key Terms
The vocabulary around AI agents fills up fast. This glossary defines the terms you will meet when reading about OpenAI Dots — in plain language, and honestly about what is still unpublished.
Last verified: September 30, 2026 · Reviewed by AstraDot editorial desk
Agent products arrive with their own vocabulary, and it is easy to nod along without ever pinning down what a word means. This reference fixes that. Most terms below are general to the industry and are described in a vendor-neutral way; the ones that are specific to OpenAI — a Dot, GPT-6 Astra, specialist Dots — are defined using OpenAI's own published description. Where a detail is not published, we say so rather than fill the gap with a guess.
If you want the story rather than the dictionary, start with what OpenAI Dots are, then read the GPT-6 Astra explainer and the security and privacy guide.
Terms, A–Z
| Term | What it means |
|---|---|
| AI agent | Software that pursues a goal by planning and taking steps, rather than only replying to a single prompt. Agents can use tools, read files, and act on the result. |
| Always-on agent | An agent that keeps working in the background between conversations instead of stopping when you close a window. OpenAI describes Dots as always-on. |
| Dot | OpenAI's always-on AI agent, powered by GPT-6 Astra. Each Dot has its own cloud computer and browser, connects to 4,000+ apps, and learns from feedback. You start with a primary Dot and give it a name. |
| GPT-6 Astra | The frontier model that powers Dots. It provides the reasoning and action-taking behind each agent, rather than being a standalone chatbot. |
| Cloud computer | A separate, isolated machine that runs in the cloud. Each Dot has its own cloud computer and browser, so it does not run on your own device. |
| Proactive research | The background work a Dot does while you are away — reading, gathering, and preparing. OpenAI says proactive research uses read-only tools. |
| Read-only tools | Tools that let an agent look at and summarise information but not change it. OpenAI says read-only research cannot send messages, change app content, or control your browser or computer. |
| Auto-review | A check that compares an agent's intended action against your instructions, your Custom Rules, and safety requirements before it proceeds. |
| Custom Rules | Instructions that set how a Dot may act: allow an action, require your approval first, or block it. They are the main lever for controlling an agent's autonomy. |
| Activity view | A record of what a Dot is doing, including background work, so you can follow progress and pause or stop monitoring when you want. |
| Specialist Dot | An organizational agent with its own identity, credentials, and access to systems of record. OpenAI is testing specialist Dots through focused enterprise pilots. |
| Plugin ecosystem | The library of app connections an agent can use. OpenAI says Dots connect to more than 4,000 apps through its plugin ecosystem. |
| Systems of record | The authoritative systems where an organization keeps its data — finance, customer, and similar platforms. Specialist Dots are given access to the ones their role needs. |
| Human-in-the-loop | A design where a person reviews or approves an agent's actions at key points instead of letting it act entirely alone. Approvals and Custom Rules are practical examples. |
| Sandbox / isolation | Keeping an agent's environment separate from other systems so its work is contained. Each Dot running on its own cloud computer is a form of isolation. |
| Salesforce of record | Informal shorthand for the specific system a team treats as the authoritative source for a category of data — for example, the CRM that is the salesforce of record for customer and deal data. It is a governance label, not the company Salesforce. |
| Multi-channel agent | An agent you can reach through several channels that share context. Dots can be reached through ChatGPT, Slack, and Teams, with texting on a limited waitlist. |
| Waitlist | A limited-access list for a feature that is not yet generally available. Texting Dots via iMessage and RCS is on a limited Pro waitlist at launch. |
| Agent governance | The policies, ownership, and controls that determine what an autonomous agent may do and who is accountable. It covers access, approvals, and auditing. |
| Microsoft Agent 365 | Microsoft's enterprise controls for managing and governing agents. OpenAI is working with Microsoft to integrate specialist Dots with governance and security controls in Agent 365. |
How these terms fit together
A few of the definitions are best understood as a chain. An AI agent plans and acts. An always-on agent keeps going in the background. A Dot is OpenAI's always-on agent, running on its own cloud computer for sandbox isolation. While you are away, its proactive research uses read-only tools. Anything that would affect your accounts passes through auto-review, where your Custom Rules decide whether it is allowed, needs approval, or is blocked, and the activity view lets you watch.
The organizational half of the chain is about scale. A specialist Dot acts for a company, using its own identity and access to the systems of record. That raises the stakes, which is why agent governance matters and why OpenAI is integrating specialist Dots with Microsoft Agent 365. Personal or organizational, the same caution holds: Dots can make mistakes, so keep a human in the loop for consequential work.
Source: OpenAI, Introducing dots and the OpenAI help center, checked 2026-09-30. Generic industry terms are defined vendor-neutrally; unpublished Dot specifics are marked as such.