Skip to main content

Web chat

The web chat is the fastest way to use Imaginne: nothing to install — just open https://ai.nnumbers.com.br/chat and sign in with your organization account. It is a full workspace — conversations organized into projects, attachments, voice input, your account's files, and memory — and runs keep going on the server even if you close the tab.

Where the agent runs

In the web chat, the agent runs on the server, not on your machine. It doesn't browse your disk: files enter the conversation when you attach them. @mentions of local paths exist only in the Desktop, the TUI, and VS Code. See Architecture and Local execution.

Sign in​

  1. Go to https://ai.nnumbers.com.br/chat.
  2. You're redirected to your organization's login. Sign in with the email and password you set from your invite.
  3. Back in the chat, your identity is tied to your authenticated session — the same login that works across every Imaginne surface. You never need to create or paste an API key. See Identity.

If your session expires, a "Session missing or expired" banner appears with a Sign in button. Click it to re-authenticate and pick up where you left off.

The layout​

The screen has two areas:

  • Sidebar (left)
    • + New chat and Search (⌘/Ctrl+K) at the top;
    • Projects — the folders that organize your conversations (see Projects & search);
    • Recent conversations — your chat list, pinned ones first;
    • the bottom navigation: Files, Memory, Integrations, and Settings;
    • your email and the Sign out button.
  • Conversation (center) — the conversation title (with the project chip, when there is one), the transcript, and, at the bottom, the message field (composer).

The theme button (sun/moon) at the top of the sidebar switches light and dark instantly. The button next to it collapses the sidebar to give the conversation more room; on mobile, the sidebar becomes a menu you open with the ☰ icon.

Start a new chat​

Click + New chat in the sidebar. A blank session is created and becomes active — the empty state offers starter suggestions you can click to fill the message field.

Each chat is automatically titled from its subject. The ⋯ menu on each conversation lets you rename, move to a project, pin, archive, duplicate, and delete — see Projects & search.

Persistent sessions​

Your conversations are saved on the server, tied to your organization and your user:

  • The active session appears in the URL as ?session=<id> — you can reload the page, bookmark the link, or open it in another tab and the conversation is restored. The browser's back/forward buttons also move between the conversations you opened.
  • Reopen an old chat by clicking it in the sidebar.
  • Hosted sessions expire after a period of inactivity. To understand the turn cycle and context management, see Sessions.

Send messages​

Type in the message field and send:

ActionKey
Send the messageEnter
Line break (without sending)Shift+Enter

Use Shift+Enter to compose multi-line requests — a list of requirements, for instance — before sending it all at once. If you'd rather Enter not send, turn off Enter sends the message in Settings.

Ask for something concrete

The more specific the request, the better the result. Instead of "make a slide", prefer "create a 5-slide presentation about the launch of product X, with a cover and a next-steps slide".

Attach files​

You can hand files to the agent in three ways: the paperclip in the composer, dragging files onto the window, or pasting straight into the message field. Attachments show up as chips with an upload progress bar before you send.

Limits and accepted types are in Attachments & account files.

Dictate by voice​

The microphone button in the composer turns speech into text: click to record, click again to stop, and the transcript fills the message field — it is never sent on its own. See Voice.

Follow the run (streaming)​

The response arrives in real time (streaming via SSE): the text appears incrementally as the agent reasons and acts.

During the run, the chat also shows what's happening behind the scenes:

  • Run details — an expandable block that lists the agent's activity (reasoning, steps, tools invoked). Open it to follow the work step by step; keep it closed for a cleaner read. In Settings, the Show run details option decides whether it starts open.
  • Tool status — when the agent runs a skill or tool, the progress appears inline in the conversation.

Code blocks get a header with the language and a Copy button.

Stop the response​

While the agent is answering, the send button becomes a stop button. Click it to cancel the running execution — the conversation records "Run cancelled." and you can rephrase your request.

Runs continue on the server​

Once you send a message, the run belongs to the server, not to your tab:

  • Close the browser, switch machines, or lose your connection — the run continues. When you reopen the conversation, the result is there.
  • Switch conversations mid-response — the previous run keeps going; you can start another conversation in parallel.
  • A colored dot next to each conversation in the sidebar shows its state:
DotMeaning
RunningThe agent is working on that conversation right now.
QueuedThe run was accepted and is about to start.
Waiting for youThe agent needs your confirmation to continue.
Done — new resultIt finished while you were in another conversation. The dot clears when you open it.
FailedThe run ended with an error.

If you start too many runs at once, you'll see "Too many concurrent runs. Wait for one to finish." — the cap is 10 active runs per user.

Artifacts and downloads​

When the agent generates a file — a .pptx presentation, a .docx document, an .xlsx spreadsheet, a PDF, or an image — it appears in the conversation as an artifact card, with:

  • an icon for the file type;
  • the file name;
  • the type and size (e.g., PPTX · 240 KB);
  • a Download button.

Images get a preview just above the card. When a run produces several files, the main result is highlighted (⭐) and the rest are grouped under "Other files from this run".

Delivered files are also kept in your account, under the Files panel — see Attachments & account files.

Integrations​

The Integrations item in the sidebar opens the same OAuth connection panel (Slack, Google Workspace, Atlassian) available in the Desktop app and the /app console. State is shared: connecting here applies there, and vice versa. See Integrations.

Settings​

The Settings item opens a side panel with:

SectionOptions
AppearanceTheme: light, dark, or follow the system.
ChatEnter sends the message; Show run details (opens the tool panel automatically).
MemoryAutomatic memory (off / suggest / automatic) and the Manage memories shortcut. See Account memory.
AccountYour email, the shortcut to Organization administration (/app), and Sign out.

Preferences live in your account, on the server — open the chat on another computer and you get the same setup.

Model​

The web chat has no model picker: it always uses the nnumbers model. If you need per-turn control over the model, use the Desktop, the TUI, or VS Code. See Models.

Sign out​

Click Sign out, at the bottom of the sidebar (or Settings → Sign out), to end the session in the browser. Your identity is disconnected; your chats, files, and memories remain saved and reappear at your next login.

What stays local-only​

The web chat covers most day-to-day work, but a few features depend on your machine:

FeatureWeb chatDesktop / TUI / VS Code
Direct access to files on your diskNo (you attach them)Yes
Per-turn model choiceNo (nnumbers)Yes
Scheduled routinesNoYes (Desktop)
Remote sessionNoYes (Desktop)
VoiceYes (record → transcribe)Yes (real-time transcription)
Attachments and artifacts kept in the accountYesFiles live on the project's disk

See also​