The chat screen
- Message area — the running conversation. Replies stream in live. A new conversation starts empty; just type your first message.
- Header (right side):
- Context meter — a percentage ring showing how much of the model’s context window this thread uses. Click it to see input/output token usage and exactly which company knowledge was auto-injected into the conversation. The ring turns amber at 60% and red at 80%; near the limit a “Context almost full — new chat” button appears, because long threads get expensive and unstable.
- Model picker — the current model, click to change (see below).
- Tools — how many tools this session can use, click to manage.
- History — opens your recent conversations.
- ＋ New chat — starts a fresh thread.
Conversation history
The History button opens an overlay with all your past conversations, grouped into Today / Yesterday / This week / Older, with a “Search conversations…” box. The list follows your active department — switch departments in the sidebar and the history follows. Hover a conversation to reveal Delete (with a confirmation dialog: deleting a thread permanently removes all its messages). Thread titles are set automatically from your first message.The composer
The input box reads “Ask anything, type / for commands…”. Press Enter to send, Shift+Enter for a new line. While a reply is streaming, the send button becomes Stop.Model picker (⌘⇧M)
Choose which AI model answers:- Auto (recommended) — the platform picks the best model per question. This is the default unless you or your admin chose otherwise.
- Search all available models, filter to Free models, see each model’s price and context length, and star favorites to pin them to the top.
Tools panel
Shows every integration this chat session may use — e.g. Gmail, Google Calendar, Google Sheets, GitHub, Slack, Notion, plus any MCP servers your company added. Toggle each one on or off for this session. The Manage connections link at the bottom takes you to the Connections page to add new integrations.Context picker
Click the ✨ Context button (or type a bare/) to open the context picker with five tabs: Skills, Knowledge, Memory, Data / Views, and Tools. Tick anything you want the assistant to work with:
- Selected knowledge and memory entries appear as removable chips above the input and are injected into the assistant’s context.
- Selected skills appear as colored chips (with Clear all when several are active).
- Selected views scope the assistant’s data queries.
- The Knowledge and Memory tabs include a + New entry button so you can create an entry on the spot — it is saved and attached immediately.
@-mentions
Type@ to mention a data view (@view_name) or a skill directly in your message. Use ↑/↓ and Enter to pick. Mentioned views and skills are enabled for that message.
Slash commands
Type a command at the start of your message:
Example:
/decide We move the launch to Q3 → the entry is classified automatically, then a confirm card appears for review before anything is stored. If you write something like “remember this” in plain language, the assistant shows a hint chip suggesting the matching command.
Confirm cards
Every save (slash command, the assistant’s suggestion, or a Save as… button) shows a card in the transcript before anything is stored. It previews Title, Intro, Class, Category, Scope, Department, Importance, and Tags, with a body preview toggle. Buttons: Cancel, Adjust (edit all fields inline), Save.
If a similar entry already exists, the card warns you before you save a duplicate. On success the card is replaced by a green ”✓ Saved as …” line.
Working with replies
- Tool calls the assistant makes (searches, mail, calendar, data queries) appear as collapsible cards with a status pill (Pending / Running / Done / Error). Expand one to see the exact arguments and result.
- Tables in an answer render as table cards with Copy as TSV, Download CSV, and Maximize (full-screen view).
- Hover a reply for Copy and Retry (regenerate).
- Under every reply: Save as Memory, Save as Knowledge, Save as Skill — each opens a pre-filled confirm card.
- After each reply a follow-up chip appears (“Suggest a next step” or “Ask a follow-up question”).
Command palette (⌘K)
Inside Chat, ⌘K opens a quick palette with actions such as New chat, Chat history, and quick prompts (today’s inbox summary, today’s calendar, knowledge search).Who am I talking to?
Chat talks to one company assistant (your admin can customize its persona under Branding). There is no per-message agent picker in Chat — your configured agents work through triggers, schedules, and missions, and report to you via the Inbox. Your active department (sidebar switcher) shapes the conversation implicitly: it filters your history, sets the default department for saves, and selects which department knowledge is auto-injected.The floating Data Assistant
On data pages (any Data view and the Inbox) a floating chat button appears bottom-right (“Query your data”). It opens the Data Assistant — a compact chat panel pre-scoped to the view you are looking at, with its own model picker and the same context picker, tables, and save buttons. Notes:- The panel can be expanded to a large view or closed with ✕.
- Conversations in the floating panel are not saved to your chat history.
- Slash commands are not available here — use the main Chat page for those.