Product features: messages, conversations, and attachments

Tell the help agent what you want done and it tries to do it; only “how / how do I” questions get a guide. See Chat as action.

Who you are talking to in Messages

• Just registered or no workspace selected: Messages defaults to the help agent. Ask how the product works, or say “help me create a workspace” (it asks for a name; otherwise My workspace). Click: Go to Messages.
• Workspace selected: pick an assistant in the left agent list, then chat on the right; you can still pick the built-in help agent for platform questions.
• After you create a new assistant, return to Messages and select that assistant before sending; My agents cards usually have a Go to chat entry.

Person-to-person chat with colleagues and friends is in Contacts (feature menu), not mixed into Messages. See contacts-and-im.md.

How conversations work

• Open Messages.
• What you send lands in a conversation; several rounds with the same agent usually share context.
• New conversation starts a new topic (click: New conversation). Wide screens show a list on the left; narrow screens open history from a menu or icon. You can pin the conversation list so the right-side work-results pane opening or closing no longer auto-collapses it.
• Take a conversation with you: tell the current assistant “export this conversation”, or in the conversation list tap ⋯ → Export conversation pack. On another server, select the agent to continue with, then say “import conversation” with the pack attached, or ⋯ → Import conversation pack. It opens as a new conversation. You take body text and attachments, not tool traces.
• Conversations and messages are tied to the current workspace, current agent, and your account. Switching workspace or agent changes the list.

How attachments work

• In the composer tap Add attachment; after upload, pending items list under the composer.
• Sending the message also sends attachments to the current agent so it can use the file contents.
• During generation you can tap Stop. While an upload or model reply is in progress, some buttons are temporarily disabled.
• For sensitive content, follow your organization’s security rules.

How many and how large per message

Allowed types: images; Word / PPT / Excel; PDF; common text; archives; and other ordinary files (for example Illustrator .ai). The assistant identifies the real type first; if built-in tools cannot read it, it tries to write a reusable skill script. Programs and installers (such as .exe / .msi) cannot be attached.

Archives have another layer: after extract, total ≤ 80MB, at most 30 analyzable files. Encrypted archives cannot be opened.

Human-support submissions use the same count and size. Over the limit, the UI shows current size vs cap.

How files appear in chat

Files you or the assistant send follow In-chat file preview in Account → Preferences:

When this conversation already has files, a preview location icon appears in the composer (bubble vs right layout). Tap to switch and write back the preference. Switching back to bubble keeps your last in-bubble mode (show / collapsed / no embed). Conversations with no files yet do not show this icon.

With View on the right and files already in this conversation:

• The right work-results pane appears; the feature menu and conversation list collapse to give width to chat and preview.
• You can still hover or tap to open the feature menu; you can also collapse work results and open Work results again later.
• Several files: Previous / Next (loops), or tap a file name; default is the latest.
• Pin the conversation list so it no longer auto open/closes with work results.
• Markdown on the right fills the pane.

Fine-tune expand behavior at Account → Preferences → In-chat file preview.

Topic of a work-agent conversation

With a workspace selected, talking to a work agent, each conversation has a topic: one or more paragraphs that state what this conversation is trying to finish, scope, and current progress. The UI shows it formatted. The short title in the list is only for recognition — not the same as the topic.

Tell the assistant; you do not need a settings page:

The top bar defaults to one-line summary; tap the bookmark on the left to edit the source, tap the summary only to expand/collapse. The help agent, website visitors, and scheduled conversations have no topic.

• Make a presentation (.pptx): Make a presentation. If you already discussed what to write, pick a style and send. Publish your own styles in the Asset library first.
• Write a document (.docx): Write a document. Can use content already in this conversation.
• Generate a PDF (.pdf): Generate a PDF. Pick a style and send; no script, no Word-then-Save-As.
• Make a spreadsheet (.xlsx): Make a spreadsheet. Colors follow the chosen style.
• Generate an image (scenery, illustration, poster — not bar/line charts): tell the current assistant “generate an image of…” or “draw…”. Use a model tagged image generation, or keep Auto. The image appears in the conversation. If one attempt fails, try again or another model that can draw. For “how do I generate an image”, the assistant only explains this wording — it will not send you to draw on the public web.

Turn a conversation into a skill or an app

After a workspace is selected, tell the help agent or a work agent:

Creating or renaming a workspace does not require a workspace selected first. Generating a skill or creating an app with no workspace selected: the help agent asks you to select one, then say it again. Do not paste the body into Skills Center yourself — say it in chat.

Work agents usually read the skill before fetching data, writing files, or running scripts. Full notes: Skills and agents.

For previewable HTML such as charts / org diagrams / store maps, see Interactive HTML report scripts (ECharts, Mermaid, Leaflet).

How the reply process looks

• The assistant’s thinking can be expanded as Markdown.
• For downloadable sheets, documents, or images, say what you want; the assistant calls office-document, write-file, or image-generation tools. If it only read a file and verbally agreed but did not generate, the system keeps prompting until a download link actually appears.

Human support

When the agent has human support enabled and someone is on shift, the shortcut bar above the composer ends with Human support. Describe the issue and submit; a local seat or authorized support group takes the ticket and contacts you via Contacts. If nobody is on shift, only Submit a ticket remains.

• Progress: Workspace collaboration → Support queue (my support requests).
• Setup and handling: Human support.

Composer tips

• Enter sends, Shift + Enter newline.
• A ring near the composer may show how much context this conversation has used; near the cap, start a new conversation.

Have the assistant remember important content

When talking to a workspace My agent (not the help agent), tap Remember on a message, or say “please remember…”. After each Q&A the assistant also auto-extracts content worth keeping long term, even if you did nothing.

Memory, knowledge base, skills, memos, and past conversations are different layers: team notes go in workspace knowledge, role metrics in agent knowledge, preferences and decisions about you in memory, records to look up later in memos. To pull original text from earlier chat (e.g. words you asked about in the last three days), tell the current assistant — see Look up past conversations. Details: Agent memory.

Related docs

• Account (identity and preferences) — default in-chat file preview
• Asset library — workspace-shared files and the same preview
• Make a presentation, Write a document, Generate a PDF, Make a spreadsheet
• Work results on the right — view on the right, make room, pin
• Conversation topic — change topic, add to topic, summarize topic