Conversation topics: state the purpose, change it in one sentence

In user language: for people who get continuous work done with the same work agent. Every conversation has a topic: one or several paragraphs that state “what we are trying to finish, the scope, and current progress”. Say you want to add a point, change it to something, or summarize, and the assistant does it; only when you ask “how do I change the topic” do you get a how-to. You do not have to memorize a fixed phrase.

For the user-facing product notes, see [docs/产品介绍.md](/docs/product-intro); for the mechanism source of truth, see [docs/core-mechanisms/对话主题与复盘.md](/docs/mech-topic-review).

Date: 2026-08-26

The takeaway (read this first)

One conversation will run many rounds. The short title in the list only helps you recognize “which one”, often a truncated first sentence. What you actually need to hold onto is the topic: the user’s main purpose in this conversation.

A topic is not a one-line tag, and it is not another truncated quote of your words. It is one or several paragraphs written in Markdown, shown formatted in the UI. Written clearly, the assistant knows what this conversation started as; each turn still follows your latest message, and the topic is not a command. When you ask something new, it answers; you only tell it to rewrite the stored topic when you want that field changed.

In one sentence: the topic records what you are trying to finish; the list title only helps you find this conversation.

flowchart TD
  A["First round ends"] --> B["Assistant drafts the topic"]
  B --> C{"What you say"}
  C -->|"change to… / update the topic to…"| D["Replace the whole thing with the new main purpose"]
  C -->|"put into / add to / write into the topic"| E["Adds to the current topic; does not write long-term memory"]
  C -->|"summarize the topic"| F["Tightens the existing text; does not remove a member"]
  C -->|"not this for now / small talk"| G["The topic steps aside; it does not lock"]
  D --> H["Topic is background; this turn follows the latest message"]
  E --> H
  F --> H
  G --> H

Who this is for, and what it solves

It only appears in work-agent, main-site conversation. The help assistant, visitors in the site widget, and scheduled conversations do not do topics.

Typical situations:

• In the same conversation you need to write characters, settle the voice, and change the outline — you need a body of “what we are actually doing now”, not a four-character title;
• You say “humour is important, put it in the topic”, and you want it written into this purpose, not treated as long-term memory across conversations;
• The topic grew long, you say “summarize, remove wording that does not fit”, and you want it to tidy the topic, not remove a colleague from the workspace.

For people getting work done: say change and it changes, say add and it adds. For people who only ask “how do I change it”: they get the steps — it will not mistakenly create a new resource.

What a topic is and is not

A common structure (the assistant can draft this way too):

• Main purpose: the thing to finish
• Scope and constraints: voice, audience, what not to do
• Current progress: what is already settled, what the next round continues

The UI renders Markdown. When the list and top bar are collapsed, the summary skips headings such as “Main purpose” and shows the first real sentence of what you are trying to finish.

What you can say to the assistant

Say what you want done, and it acts; ask “how / how do I” and it explains.

Adding to the topic is not remembering. “Remember” writes long-term memory, which still holds across conversations. Adding to the topic only changes this conversation’s purpose, so “this book should be humorous” is not written as “every future conversation should be humorous”.

Tidying the topic is not removing a member. When “remove” sits next to “wording / content / topic”, it is handled as changing the topic. Invite, remove a colleague, delete an agent — those change-or-delete actions look at the full sentence first, so “remove wording that does not fit” is not blocked as an HR operation.

On a topic change, the assistant asks first whether to change; you nod, then it changes. The topic is an anchor, not a lock.

How to see and change it in the UI

1. Top-bar topic strip (Web): one-line summary by default. The bookmark icon on the left is the only edit entry; tap it to change the source. Tapping the summary or the arrow on the right only expands or collapses, so you do not enter edit by mistake.
2. List and collapsed state: you see the sentence of “what we are trying to finish”, not the words “Main purpose”.
3. Phone: one line of plain-text summary, without taking the composer.
4. Every round sent to the assistant: carries the current topic. When it drifts, it asks whether to change topic rather than silently rewriting.

The topic cap is about four thousand characters — enough to state purpose and progress, not a place for a long work’s body.

How it splits with review

The topic covers “what we are doing now”. Review covers “this stretch is done, what is the conclusion”.

By default it will not nudge you to review. If you turned on Prompt for review in the agent settings, after at least two rounds, when the assistant judges the work in the topic is done, Review appears in the composer or menu (it does not start writing on its own); if this conversation has been idle more than 2 hours and less than 2 days, the assistant invites once in the same conversation. If it is off, say “review this” yourself. After you edit the five columns, confirm: what was finished, key decisions, your requirements and corrections, what is still open, reusable practices.

A confirmed review can be a reference for later similar topics, not gospel. Rolling compression of conversation history is not review; the two are separate.

When not to lean on the topic alone

• You have already switched to a completely different piece of work — change topic or start a new conversation; do not mash two large pieces into the same topic.
• Preferences you want every future conversation to follow — say “please remember”; do not only add them to the topic.
• Asking the help assistant how to use the product — that conversation has no topic and does not do review.

Purpose goes in the topic, habits across conversations go in memory, team metric goes in the knowledge base.

Related reading

• Conversation review — how to review after the work is done, quiet by default, when to turn on prompts
• Conversation topics and review — the full convention for draft, topic change, add, and review invite
• Product introduction — where Messages and work agents sit in the product
• UI and layout — how the top-bar topic strip and list summary are shown
• Agent memory — why “remember” and adding to the topic are not the same thing
• Work results: view files on the right — purpose goes in the topic, files are viewed on the right