Chat & Messaging
One unified chat workspace for everything — human conversations, workspace channels, and AI sessions. Pick a model per conversation, switch between text, voice, and avatar, and pull in exactly the context you need.
One workspace, many conversation types
The Chat area brings together several kinds of conversation in a single Gemini-style interface with a history list and a "New chat" button:
- Human conversations — direct messages and workspace channels with your teammates.
- AI chat sessions — private, ChatGPT/Claude-style conversations with the assistant, kept in a searchable history with archiving.
- The persistent Assistant — a global assistant available across the app (the dedicated Assistant tab now lives inside Chat).
- Feature-design chats — edit a workflow or agent draft conversationally, right from chat.
Text, voice, and avatar
Every conversation can run in one of three input modes, chosen from the composer and remembered per session:
- Text — the standard composer.
- Voice — speak to the assistant; your words are transcribed and answered out loud, with the same tools and context as text.
- Avatar — voice interaction alongside an animated avatar tile (powered by Simli).
Because voice is routed through the same chat engine as text, switching mid-conversation keeps full context. See Unified Voice + Chat for how the voice pipeline works.
Choosing a model
Each conversation has its own model selector with your recent and effective models. Switch providers — OpenAI, Anthropic, Google, Groq, Ollama, or Ollama Cloud — at any point. Every AI reply records the model that produced it, so you always know what answered.
Adding context with #
Type # to open the context picker and attach live context to your message. Depending on where you are, you can pull in:
| Context | What it adds |
|---|---|
| Agent | Delegate the turn to a saved AI agent with its own tools and knowledge. |
| Prompt | Insert a reusable prompt template from your Prompt Library. |
| Knowledge folder | Ground the answer in a specific knowledge folder. |
| Integration | Give the assistant access to a connected integration. |
| Custom integration | Use one of your user-defined HTTP integrations. |
| Skill | Switch on a reusable instruction snippet for this chat. |
| Memory | Reference a stored memory key-value pair. |
| Workflow | Bring in a saved workflow for the assistant to reference or trigger. |
Type @ to mention people or the AI. Mentions are stored as structured references, not just text.
The tool strip
Below the composer, a tool strip exposes the assistant's extra capabilities. Each tool appears only when it's available in your organization:
- Web Search — let the assistant search the web for current information.
- Image Generation — generate images from a prompt.
- Data Analysis — analyze CSV or tabular data and generate charts or statistics.
- Canvas — open a side-by-side document the AI writes and edits.
- Deep Research — kick off a multi-step research task.
- Knowledge — attach knowledge folders to ground the conversation.
Side panels
Canvas and Deep Research open beside the chat rather than replacing it — a split view on desktop, an overlay on mobile — so you can watch a document take shape or follow a research plan while you keep chatting.
Watching the AI work
When the assistant uses a tool, its tool calls are visualized inline as cards (and in a side panel), so you can see exactly what it searched, fetched, or generated. You can also set per-chat custom instructions and open a Memory sheet to manage what the assistant remembers.
Search, translation, and undo
A cross-chat search lets you find messages across all your conversations. Any message can be translated on demand, and a spoken message carries a badge distinguishing it from a typed one. Just sent something you'd like back? Undo send gives you a short window (about 30 seconds) to retract a message.
File attachments
Attach files to a message, or attach a knowledge document or an integration-action field directly as context. Attachments become part of the conversation the assistant can reason over.
The chat experience is built from an embeddable dock that's reused across the app as a panel, dialog, or inline component — so you can ask the assistant for help from many screens, not just the Chat page.
Your Office AI's chat is focused on conversations and AI assistance. There are no emoji reactions, threaded replies, or message editing — if you need to revise something, send a follow-up or use Undo send right after posting.