v0.18.0

Backfilled release page. Originally dated 18 July 2026 in the changelog; the tag is the record.

Added

• Top memory users in pressure warnings — the memory warning alert now lists the top 5 memory-consuming containers across all running containers (largest first), not just the killable ones, so you can see at a glance who is actually eating the RAM. One bounded snapshot (_memory_snapshot, capped by stats_timeout) feeds both the text and the button labels; if Docker is slow under pressure the alert falls back to names without figures rather than being delayed.
• Memory restart list (memory_management.restart_containers) — containers that hog memory but recover after a bounce (classic Plex) can be offered a one-tap "🔄 Restart" button on warning and critical memory alerts, as the gentle alternative to stopping. Restarting the container that is mid-countdown for an auto-kill cancels the countdown. Restart buttons are rejected for protected containers and for names not on the list, even if spoofed.
• Configure memory restarts from Telegram — /manage → ⚙️ Features → 🧠 Configure memory restarts opens a tap-to-toggle container picker (same UX as auto-heal). Saves persist to config.yaml and apply live — the running MemoryMonitor shares the config object, so no bot restart is needed.

Changed

• Stop/Restart buttons on memory warnings are sorted by memory usage, largest first — the biggest win is always the top button. The auto-kill order is unchanged and still follows the configured killable_containers priority order.
• Memory alerts fall back to plain text if Markdown parsing fails — container names can contain Markdown special characters (underscores are common); a parse failure now resends the alert without formatting instead of silently dropping it during a memory event.

---

Full history in CHANGELOG.md.