Ask a conversation to save itself as a note: the assistant files it in the right folder and writes it in the background, and every tool it uses is now listed under the reply.
8 September 2026
Until now, a conversation and your notes lived side by side and never touched: if a thread worked something out, copying it into a note was your job. In 1.54.0 you can just ask. Tell the assistant to save what you settled on and it creates the note, files it, and writes it while you carry on. The rest of this release tidies up how a reply shows what the assistant did along the way.
Save a conversation as a note by asking for it. Say something like "save this as a note in Travel, just the itinerary" and the assistant creates the note for you. Whatever you say about what the note is for and what to keep or leave out is passed on as written. It only does this when you ask. It never saves a note on its own initiative.
Filed where you'd put it. The assistant sees your folder list, including which folder sits inside which, so naming a folder files the note there. It reuses an existing folder whenever one plausibly matches and creates a new one only when none does. If you don't name a folder, the note goes to the top level.
Written in the background, straight from the thread. The note appears right away as a card in the conversation and fills in once it's written. The content is a short, standalone summary of the conclusions: no greetings, no dead ends, no "the assistant explained that…", and no links the conversation didn't contain. While it's being written, the card reads Writing in the background — ready in your sidebar shortly. When it's done, the card shows the folder it landed in, or Top level. If writing fails, it says Couldn't write this note — open it to retry.
Open the note from the card. Tapping the card on your phone opens the note. On desktop it opens in a new panel, or switches to that panel if the note is already open. If all your panels are in use, you're told how many you can have open instead of losing one.
Tool activity grouped under the reply. Anything the assistant did while answering is collected in a collapsible 1 tool call / N tool calls section under the reply. It's open while the reply is still generating and folded away afterwards. Web searches appear there as Searching the web / Searched the web, with a result count that expands into the list of pages consulted. This replaces the separate sources list.
Calmer replies while they generate. Retry and Copy now appear only once a reply has finished. Before, they showed the whole time it streamed, with Retry disabled. The Thinking panel has a lighter look and is part of the reply itself.
Long messages collapse. Your own long messages are trimmed to a few lines with Show more / Show less, so a pasted wall of text doesn't push the conversation off the screen.
A tighter composer. The message box starts at one line and grows as you type. On phones, Enter adds a new line instead of sending, and the conversation's empty-state panel steps aside while you're typing.
Kimi K2.6 is the new default model for conversations that haven't picked one.
A proper "note not found". Opening a note that was deleted or never existed now shows Note not found with a way out: Close in a panel, Back on a full page. Before, you got an empty editor.