Editing your message and regenerating

Hover over any of your messages, click the pencil icon → change the text → Send.

Short answer

Hover over any of your messages, click the pencil icon → change the text → Send. The old model response and everything after it get deleted, and the model generates a fresh reply to your edited version. Like ChatGPT’s edit-and-regenerate.

Detailed explanation

Where the button is. Hover any user message (not assistant messages) - action buttons appear on the right: Copy · Edit · Resend · Delete. Edit is the pencil icon.

What the dialog shows. A modal with a multi-line text box pre-filled with your original message. Same visual style as the main chat input. Auto-height with a minimum of 3 lines and scrolling for longer text. Buttons: Cancel · Send.

What happens on Send.

  1. Your message content updates to the edited version (in memory and on disk)
  2. All messages that came after it in this chat are deleted - both assistant replies and any subsequent user turns
  3. The model generates a fresh answer to the edited message
  4. Streaming behaves normally: pulsing cat, cancelable, etc.

On Cancel. Nothing changes. Dialog closes, chat continues as before.

Only for user messages. Assistant messages don’t have an Edit button - you can’t rewrite the model’s response. If you want a different answer from the model, use Resend (same message again) or Edit your prompt.

Why it matters

The common pattern: you ask something, the model misunderstands or you realize your question was ambiguous, you want to ask better without starting over. Edit-and-regenerate solves this without cluttering the chat with corrections. Especially useful for prompt engineering - tweak wording, see how the answer changes.

The “delete everything after” behavior is important: it prevents the model from being confused by seeing its own now-outdated reply in the context. The chat effectively branches at the edited message, and the old branch is dropped.

In Abliteration specifically

Edit dialog matches the neko-theme dialog style used elsewhere in the app (consistent buttons, colors, spacing).

Max length matches the chat input’s max - no artificial cap. Same font, same wrap behavior.

Deleted subsequent messages are gone - no undo. If you edited by accident and want to recover, closing without saving the app before the edit was the only way, and even that’s not reliable. Treat Edit as a commit.

If you edit a message that used Council mode (multiple model replies after it), all Council replies get deleted too, and regeneration continues in whatever mode is currently active (Council on/off toggle state).

Catalog is the map. Apps are the tools.

Run models on your own machine, not in the cloud.

Every model page has an "Open in Abliteration" button that hands the model directly to the first-party desktop client, at the quantization your rig can actually run. No API keys, no subscription, no prompt leakage.

Open in Abliteration