Chatting with Your Assistant

Chatting with your assistant feels like talking to a knowledgeable colleague. You type naturally, ask follow-up questions, and the assistant responds with context-aware answers. No special syntax or commands required.

Starting a conversation

Click Assistant in the sidebar to open a new chat. You can also resume an existing conversation from your History. Each conversation is its own thread — you can have multiple going at once without them getting tangled.

The capability picker

On a fresh thread (no messages yet), the empty state shows a Recommended for you grid of capability cards — quick entry points like "Draft an email", "Summarize a meeting", "Research a topic". Click any card to drop its example prompt into the composer. The cursor lands at the end so you can edit before sending; the message is not sent automatically.

The cards reorder based on what you've used recently — the capability you clicked most recently appears first the next time you start a fresh thread.

The slash menu

Inside an existing thread, type / at the start of a line in the composer to open an inline slash picker. It groups commands as:

• Recent — up to 3 capabilities you've used recently
• Capabilities — the full catalog

Type after the / to filter. Pick an item with arrow keys + Enter (or click). The /query token is removed from your message and the capability's example prompt is inserted in its place.

This is the fastest way to invoke a known capability without scrolling through the hero or the dialog.

Agent and Plan modes

Next to the composer is a small Agent / Plan toggle that controls how the assistant responds to your next message:

• Agent — the default. The assistant executes tasks with full tool access: creating items, editing documents, scheduling, and anything else it can do.
• Plan — a read-only mode. The assistant researches and proposes a plan but makes no changes until you switch back to Agent and approve. Use it when you want to think through an approach before anything happens to your workspace.

The toggle is per message, so you can plan first, review the proposal, then flip to Agent to carry it out in the same thread. The same toggle appears on the home start page, so you can pick a mode before your first message — your choice carries into the new thread.

Picking a model

The model selector beside the composer lets you choose which AI model answers. v1.0 supports 11 providers across US, EU, and Asia regions — pick based on:

• Reasoning depth — larger models (Opus-class, GPT-class large) handle complex synthesis better
• Latency — smaller models (Haiku, Sonnet, GPT-class small) respond faster
• Region / privacy — pick a regional model if your team has data-residency requirements

Your choice persists per thread. The default model is set by your workspace admin.

Types of messages you can send

Your assistant handles a wide range of requests:

• Questions — "What's the status of Project Alpha?" or "How does our refund policy work?"
• Writing help — "Draft an email to the team about the schedule change" or "Help me outline a blog post"
• Analysis — "Compare these two approaches and list the pros and cons" or "Summarize this document"
• Brainstorming — "Give me 10 name ideas for our new feature" or "Help me think through this problem"
• Research — "What are the best practices for onboarding?" or "Find relevant info in our knowledge base"
• Actions — "Create a task for the marketing team" or "Schedule a reminder for Friday"
• Voice messages — Tap the microphone to speak instead of type. Your speech is transcribed in real time, or start a full voice conversation for a hands-free experience.

How the assistant uses context

This is what makes your assistant different from a generic chatbot. When you ask a question, it draws on multiple sources:

• Current conversation — Everything you've said in this chat thread
• Memory — Key details from past conversations (your preferences, decisions, context you've shared)
• Knowledge base — Documents you've connected, so answers come from your actual content
• Uploaded files — Any files you've attached to the current conversation

The more context the assistant has, the better its responses. That's why connecting your knowledge base and letting memory build up over time makes such a difference.

Follow-up questions

You don't have to start a new conversation every time you want to dig deeper. Just keep asking. The assistant tracks the full thread and understands references like "Can you expand on that?" or "What about the second option?"

This makes it great for iterating on ideas. Start with a broad question, then narrow down with follow-ups until you get exactly what you need.

Tips for better responses

• Give context upfront — A sentence of background helps the assistant tailor its answer
• Be specific — "Summarize in 3 bullet points" is better than "Summarize this"
• Iterate — If the first response isn't quite right, tell the assistant what to change rather than starting over
• Reference files — Upload or mention specific documents so the assistant works with real data, not assumptions
• Set preferences — Tell the assistant how you like things formatted. It'll remember for future conversations.