Console Operator
The Console Operator—build and run work from one chat, scoped to a project.
The Console Operator is the chat surface in Console. Describe your goal, and it plans the work, hands each part to the right agent, and returns the result in the same conversation.
Use it to build and use apps, agents, knowledge bases, and agentic applications. It can also search the web, crawl a website, answer questions about data, generate documents, and run on a schedule.
It replaced the Console dashboard. The dashboard listed existing assets. The Operator gives you one place to ask and act.
For general orchestration, see Operator. This page covers Console behavior. For every task in one place, see Use the Console Operator.
At a glance
Where: Operator in the Console sidebar, in your selected project
Scope: Project apps, agents, knowledge bases, design systems, and context
Acts: immediately. It dispatches when you ask—there is no approval step
Also builds: apps, agents, knowledge bases, and agentic applications
Searches: the web and public websites
Runs on its own: on a schedule you set. See Schedules
Waits for you: when an agent needs an answer. See Gates
Start a conversation
The Operator opens on a greeting with your name, such as "Good evening, Abhishek", and the subtitle "Ask a question or tell me what you need, I'll bring the outcome."

Type your request in How can I help you today? and send it, or pick a starter from the row under the composer, such as Approval Flow or Review a Policy. Arrows either side of the row scroll through the starters.
Recent chats lists your latest conversations, each with a time such as 5 Hrs ago. View all chats opens the full list.
Work in a conversation
A conversation has three columns.

Conversation History, on the left. Your conversations in the project, newest first. Conversations are specific to each user, so you see only the ones you started. Search conversations filters them by title.
The transcript, in the middle. The Operator shows its work as it goes: how long it thought, each step it took (such as Update Plan Steps or List Database Providers), and each question it asked with the answer beneath it. When it hands work to an agent, the agent appears as a card with its step count, for example App Builder Agent · 49 steps, and a Working status while it runs. Select Open full run on the card to see every step the agent took. Jump to latest returns to the bottom. + New Conversation, top right, starts over.
The side panel, on the right, in four sections:
Agents — the agents the Operator can hand work to, under System and User. A count shows how many are working now, for example 1 working. See Agents & connections.
Connections — the Integration Hub connections the Operator can use, and All integrations. See Agents & connections.
Outputs — what the conversation built, such as an app, with a link to open it.
Artifacts — files and documents created during the task: "Artifacts created during the task land here."
When an agent asks you something
Agents stop to ask when they need a decision, for example "There are two apps named ToDo List App in your project. Which one would you like to update?" The question appears as a card above the composer. Pick one of the numbered answers, write your own in Answer in your own words…, or select Skip. When the agent asks for a file, such as a brand guide, attach it as your answer. A conversation that ends on a question also appears on the Gates page until you answer.
Long-running work
Building an app or ingesting knowledge can take several minutes. The Operator tracks these builds as background tasks, so it reports the real outcome when the build finishes, even if the conversation was interrupted. You may see "Resuming 1 interrupted tool calls…" when it picks one back up.
When a turn runs long, a banner appears above the composer:
This is taking a while "You can pick this up in a new conversation with a summary of where things stand. The work here keeps running either way."
Select Continue in a new conversation to carry on elsewhere with a summary, or Stay here to keep watching. The build keeps running whichever you choose.
The composer
The composer at the bottom holds your message and four controls:
+ — opens Attach files, to add files to your message, and Connections, to turn the Operator's connections on or off. See Agents & connections.
Model — the model the Operator runs on, for example Gemini 3.7 Flash or GPT 5.4.
Effort — how hard the Operator reasons before it answers: Minimal, Low, Medium or High. More effort usually gives better plans on complex requests, at the cost of time and credits.
Context gauge — how full the conversation's context is, for example 0%. Hover over it for the detail, such as 18 / ~200.0K tokens. When it fills up, start a new conversation.
Your model and effort choice applies to each message you send, so you can switch mid-conversation: plan on one model, then build on another. The composer remembers your last choice for your next conversation.
The choice applies to the Operator only. Agents it hands work to, such as the App Builder Agent, run on their own configured model and effort.
What it does
It routes work to your agents. The Operator interprets your intent, identifies the agent that can complete each task, and combines its updates into one response.
It builds. Ask for a knowledge base, app, agent, or agentic application. The Operator creates it in the current project. You can test a new agent in the same conversation. When you ask for an app, the App Builder Agent asks which of the project's design systems to use, or you can name one in your request. If the branding doesn't exist yet, it saves a new design system as part of the build.
It uses the web as input. Ask it to search the web or crawl a public website. Given a URL, it can use the site's content to create a knowledge base. It can also use the website's design system when building an app.
It works with documents. The Document AI Agent extracts data from documents, classifies and summarises them, answers questions from them, and generates new ones. It can save a document template—sections and their order, layout, and a theme—and generate later documents from it. A document generated from the same template twice has the same structure and look, and a design system saved with the template keeps its branding without being named again. Before it generates a document, the agent lists the project's saved templates and asks you to choose one; it never picks one for you.
It manages design systems. Ask it to create, change or delete one of the project's design systems, for example from a brand's website. To set a logo, attach the image to the conversation. When it generates a document, it asks which design system to brand it with, and the document takes that design system's colours.
It uses your connections. With a connection turned on, the Operator can act in that third-party system. When a request needs a system it has no connection for, it tells you which existing connection to turn on, or which integration you could connect. See Agents & connections.
It helps you look after the account. Ask who is in the account and which teams it has. The Operator can tell you, but it can't add, remove or change people or teams; that stays in Users and Teams. It can also manage the account's projects for you.
It works inside one project. Everything it creates belongs to the selected project. Switching projects changes what the Operator can reach. See Accounts, Projects, Spaces for these boundaries.

What you control, and what you don't
The System agents the Operator dispatches to are preconfigured for the account. You don't assemble that team, and you can't edit its agents from the Operator. The set is expected to grow as more capabilities ship.
What you do control is which agents and connections the Operator may use: turn a system agent off, turn your own project agents on, and choose which connections it can reach. See Agents & connections.
For common questions and fixes, see Console Operator: FAQs & Troubleshooting.
Related
Maintained by Abhishek Paul
AI-assisted, human-approved

Last updated
Was this helpful?

