Chat
Chat is the direct line. No agent, no schedule — you ask, it answers, and it can act while you watch.
What it can reach
Section titled “What it can reach”Your knowledge base. It reads notes on demand, only on the turns that need them, and it can write too — creating and editing notes as you talk. Ask “what did I decide about pricing?” and it finds the relevant notes and answers from them.
Your connected accounts. Every active connection in the workspace is available. “Check if I have anything from the accountant” works without building an agent for it.
Your context. Your profile and everything in memory/ is always present, so
it knows your timezone and how you like to be spoken to.
The web. It can search, fetch a page, and — when a page only renders with JavaScript — open it in a real browser and read what actually appeared.
What it cannot do
Section titled “What it cannot do”Chat is deliberately narrower than an agent:
- It can create and edit notes, but not delete or rename them.
- It cannot run shell commands.
- It can read a page in the browser but never click, type or sign in — that belongs to agents you have granted it to, for the same reason as below.
Agents get the wider set because they are reviewed and tested before they are saved. A chat turn is not.
Where to use it
Section titled “Where to use it”The web interface — full-page chat, or the slide-over from anywhere.
A chat app — type anything that is not a command into Telegram, Discord or
Slack and it becomes a chat turn. /chat manages conversations explicitly:
start, list, resume, stop, delete.
Both write to the same conversations.
Conversations
Section titled “Conversations”Every chat is saved and resumable. They are also written into the knowledge base
as readable markdown under chats/, so a conversation you had three weeks ago is
searchable alongside your notes.
A conversation stops on its own after 30 minutes of silence. Reopening it from the interface resumes it automatically — there is nothing to restart.
Leaving mid-answer
Section titled “Leaving mid-answer”A question can take minutes to answer, and you do not have to sit and watch it. The turn runs on the server, not in your browser tab: your message is saved the moment you send it, and the answer is saved when it arrives. Close the tab, walk away, come back on your phone — your question is there, and so is the reply if it has landed.
Come back while it is still working and you see what it is doing right now — the file it is reading, the search it is running — with the earlier steps one click away. Only one turn runs per chat at a time, so sending a second question while the first is still working is declined rather than queued.
Approvals
Section titled “Approvals”Chat is not gated by the approval system that agents can use. When you ask it to send something, it sends it.
That is deliberate: you are present, and holding a live request for approval would mean holding you against yourself. The trade-off is that you have not reviewed the exact wording before it goes — so for anything public and irreversible, an agent with approval turned on is the safer route.