> For AI agents: the complete documentation index is available at /llms.txt, the full documentation bundle is available at /llms-full.txt.

# 2026-10-24

## Improvements <Badge type="warning" text="Improvements" />

- **Smoother "More (n)" popover**: opening the "More (n)" entry that holds collapsed conversations in the home sidebar used to stutter with many collapsed conversations (two or three hundred) — every row rendered at once the moment it opened. The popover now uses virtual scrolling and renders only the rows near the viewport, fixes a first-frame flash where it briefly rendered at full height before positioning, opens instantly, and still scrolls through every collapsed conversation (HF-14500).
- **Sidebar conversation list capped at 30**: the home sidebar conversation list now shows at most 30 recent conversations outside "More"; older ones collapse into "More (n)" automatically, so a long list no longer fills the screen (HF-14500).

## Fixes <Badge type="danger" text="Fixes" />

- **"More (n)" no longer jumps to the top when scrolling up**: after opening the popover, scrolling up with the wheel occasionally jumped the list straight back to the top — when a focused row left the virtualized render window, focus management restored focus near the top and dragged scrolling with it. Focused rows now stay in the render set, and scrolling up moves smoothly row by row (HF-14500).

- **Independent inheritance for duplicate Runtime settings**: in a duplicate's Runtime settings, the default model, environment variables, custom arguments, and timeout now each inherit independently — editing environment variables no longer flips the default model to "overridden", and restoring the model's inheritance no longer resets environment variables. Each item has its own "inherited / overridden" badge and "restore inheritance" entry, and changes to the parent's settings still sync automatically to items you haven't overridden (HF-14509).