Skip to content

[Feature Request] Quote & Reply: comment on any selected part of an AI response in Kimi Web #2601

Description

@topit

I'd like Kimi Web to support quote-and-reply on assistant messages: let the user select any text span from an AI response (a paragraph, a code block, one step in a plan, a line in a diff explanation), attach a comment or follow-up question to exactly that selection, and have the agent continue the conversation with the quoted fragment as explicit context.

Today the conversation is strictly linear — the only way to respond to an AI answer is to write a new message at the bottom and manually copy-paste the part I'm referring to. For long outputs (plans, multi-file diffs, research summaries) this breaks down quickly:

  • Imprecise corrections: when I want to push back on one specific claim or one step of a plan, I have to re-type or paste the fragment and hope the agent locates it.
  • No per-item discussion: an AI answer often contains several independent decisions. There is no way to approve item 1, question item 3, and reject item 5 — it's all-or-nothing.
  • Lost context in review flows: when reviewing a plan or a diff explanation, I want the agent to see exactly which sentence my comment is about, not my reconstruction of it.

Prior art worth looking at:

  • Craft Agents — any part of an AI response can be selected and commented on ("追问" / follow-up); comments stay anchored to the selected fragment and become quote cards that are sent with the next message, so the agent knows exactly what each comment refers to. (Screenshots in "Additional information" below.)
  • ChatGPT — selecting text in a response offers "Quote" and inserts the fragment into the input box as a reference block.

Proposed behavior (minimal viable version):

  1. User selects text inside an assistant message in Kimi Web.
  2. A small affordance appears (e.g. "Quote & reply" / 💬).
  3. The selection is inserted into the input box as a quoted reference block; the user types their comment below it.
  4. On send, the agent receives the quoted fragment + the comment, and the resulting exchange is rendered with the quote visually anchored to the original message.

Later iterations could support multiple anchored comments per message, comment resolution states, and "apply comments then regenerate" batch flows — but the simple quote-reply loop above already covers the core value.


Additional information

Screenshots from Craft Agents showing the interaction loop (1: select a fragment in an AI response and attach a follow-up; 2: the selection stays highlighted with an anchor marker, and a quote card is queued in the input area):

Image Image

Related requests:

I searched existing issues/discussions and did not find a request for quote/comment-based reply on AI responses; opening this to track it.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions