Chat

Chat is the main surface you land on: a prompt composer with your recent conversations close at hand. It’s where you ask questions and start and continue threads with the general assistant. The left sidebar is your main navigation. Search sits at the top. Chat is the headline destination. Click > More to open Applications, which includes Messages (pinned in the sidebar by default), Contacts, and Knowledge Manager. Settings lives in the account menu at the bottom.

​

What you can do

• Ask and converse. Type a prompt and get a streamed reply. Conversations are kept as threads you can return to. Each answer shows which tools were selected and run.
• Follow the model’s thinking. While a reply is generating, Chat shows the thought process so a long wait is progress, not a blank pause. See Thinking as the model works.
• Read earlier messages while a reply writes. Scrolling up does not yank you to the bottom. See Reading while a reply streams.
• Open code and HTML from a conversation in a side panel next to Chat, see Code and HTML in a side panel.
• Use Auto mode. Let Chat pick a persona by matching your prompt to persona descriptions, see Auto mode below.
• Search the app. Use Search at the top of the left sidebar, or press ⌘K, see Global search below.
• Use voice input. Click the microphone in the prompt window to dictate your message, see Voice input below.
• Turn web search on or off. Click + in the prompt window and use the Online toggle, see Offline and Online mode below.
• Apply a skill. Click + in the prompt window and choose a skill from the + menu to shape how a run behaves.
• Use a knowledge base. Click + in the prompt window, open Knowledges, and select a base, see Knowledge Manager.
• Schedule recurring AI work. Click + in the prompt window and choose Scheduled tasks to automate saved prompts on a timetable, see Scheduled Tasks.
• Run locally in Offline mode. When you use a downloaded model in Offline mode, replies are generated on your machine, not in the cloud.
• Manage a reply. Regenerate an answer, or use Answer as to have a different persona respond to the same message.
• Pin chats. Keep important threads under Pinned in the sidebar, see Managing chats below.
• Manage chat history. Open the full Chats list to select and delete threads one at a time or in bulk.

​

Answer as, choose which persona responds

Every reply has an Answer as control in its toolbar (the dropdown next to the regenerate and copy buttons). Open it and pick any of your personas, the General Assistant, a specialist you’ve built, or any other, and that persona answers the conversation. It’s the quickest way to get a second perspective, or to hand a question to a more suitable persona, without starting a new thread.

​

Offline and Online mode

Click + in the prompt window and use the Online toggle to turn web search on or off. The setting stays until you change it. The + menu also holds Skills, Knowledges (knowledge bases you manage in Knowledge Manager), and Scheduled tasks (recurring automation, see Scheduled Tasks). When Online is off, Chat runs in offline mode (the default). Chat uses your local setup, when you have a downloaded model, replies are generated on your machine. When Online is on, webAI searches the internet for current information and weaves the results into the answer. Replies stream as they are generated. Citations to the source websites appear in the reply so you can see where the information came from. Web search in Online mode works on entry-level machines. The Online controls also include Connected tools, which lets the assistant use your linked accounts and MCP servers (Google, Slack, and custom servers). Web search and Connected tools are independent: you can use either or both. See Connectors to connect services and use them in chat.

​

Streaming replies and tools

Replies stream as they are generated, including in Online mode. While an answer is running, Chat shows which tools were selected and which ones ran, so you can see progress and what the model actually did.

​

Thinking as the model works

While a reply is generating, Chat shows the model’s thinking so a long wait is real progress instead of a blank pause. The thought process scrolls along with the model. When the answer arrives, the thinking tucks away. You can reopen it if you want to read it again.

​

Reading while a reply streams

If you scroll up to read earlier messages while a reply is still being written, Chat does not yank you back to the bottom. Stay where you are until you choose to scroll down again.

​

Complete and interrupted replies

Long answers are not cut off partway through. If a reply is interrupted, you keep the text received so far, plus a notice that the reply was interrupted. The chat does not stay stuck on Thinking. Answers from connected personas are saved as they arrive, so a partial reply is not lost if the stream stops. See Personas.

​

Code and HTML in a side panel

Code and HTML in a conversation can be opened in a side panel next to Chat. Use the View / Preview toggle to switch how you look at the content. Preview renders styled, interactive pages. Documents and artifacts no longer appear twice in the thread. Use Show in Artifacts to open one in Artifacts.

​

Auto mode

When you send a message in Chat without choosing a specific persona, Auto mode picks one by matching your prompt to your persona descriptions.

• Your message appears in the thread immediately while routing runs, so the conversation does not sit on a silent pause.
• Short greetings get an instant reply without waiting on a full persona match.
• If matching is slow, unavailable, or unsure, Chat falls back to a faster rule-based router so you still get an answer.
• When you edit or delete a persona, routing updates right away, so a removed persona cannot capture the next turn.

To chat with a specific persona instead, use Answer as on a reply, or open the Personas tab and click a persona card. Keep persona descriptions accurate so Auto mode can match them well. See Personas.

​

Skills

Skills are reusable add-ons that shape how the assistant handles a request, applying one tailors the way that response is approached, so you don’t have to spell out the same instructions every time. To use one, click + in the prompt window and select a skill from the + menu. It applies to that message, then the assistant replies with the skill guiding how it works.

​

Voice input

Click the microphone in the prompt window to speak instead of type. webAI listens and transcribes your speech into the composer. When the text looks right, click Send to submit the message. The same microphone control is available when you chat with a persona on the Personas tab, see Personas.

​

Global search

Use Search at the top of the left sidebar to find content across the app. Click the Search bar, or press ⌘K, then type what you’re looking for and open a result to jump there.

​

Managing chats

Recent threads appear under History in the left sidebar. History is tidier so recent threads are easier to scan. When you pin a chat, it also appears under Pinned above History. Apps you pin from Applications appear there too. Direct and group conversations with people stay out of this list. Those live in Messages.

​

Pin a chat

1. In the sidebar under History, open the ⋯ menu next to a chat.
2. Click Pin to sidebar.
3. The chat appears under Pinned.

To unpin, open the ⋯ menu on a pinned chat and unpin it the same way.

​

See all chats and delete in bulk

1. On the sidebar next to History, click See all to open the full Chats view.
2. Select individual threads, or select a whole group such as Today, This week, or This month.
3. In the selection bar, click Clear to deselect, or Delete to remove the selected chats.
4. Confirm the deletion when prompted.

You can also rename or delete threads from the sidebar.

​

Common questions