Intro

I heard the news about the OpenClaw 2.22 update for my AI assistant’s core. Naturally, I applied it immediately!

If the previous 2.21 update was about giving the assistant a new brain (Gemini 3.1), this 2.22 update is all about “sharpening its memory, making it self-evolving, and turning security into a fortress.”

As a self-proclaimed “non-coder,” I’ve extracted the most practical and noticeable features from the massive changelog so you don’t have to read through all the technical jargon. Let’s dive in! 🏃‍♂️💨

1. 🧠 Multilingual Memory (FTS) Supercharged!

This is a massive win for non-English users. Previously, asking the assistant to recall specific memories in languages like Korean or Japanese was sometimes clunky due to complex grammar structures.

With 2.22, OpenClaw introduced stop-word filtering and particle-aware keyword extraction for Korean, Japanese, Spanish, Portuguese, and Arabic! (It even understands mixed Korean/English words now). Your assistant’s memory recall just got incredibly sharp.

2. 🔄 Self-Evolving Assistant: Auto-Updater & Dry-Run

Tired of manually typing commands to update your agent?

  • Built-in Auto-Updater: By enabling update.auto.*, your assistant will automatically and quietly update itself during stable rollout windows.
  • Safe Preview (--dry-run): For those worried about breaking their configs, the new openclaw update --dry-run command lets you preview exactly which files and plugins will change without actually installing anything.

3. 💬 Synology Chat Support & Cleaner UI

Shoutout to the Synology NAS users!

  • A native Synology Chat plugin is now officially supported, complete with webhook ingress, direct messaging, and media support.
  • Cleaner Chat UI: Sometimes the assistant’s replies would leak ugly system tags like [[reply_to_current]] into the chat. The new update automatically strips these control tags from the user interface. Plus, chat rendering is much faster now since it doesn’t need to refresh the entire history for every message.

4. 🌐 Browser Relay Stability & Mistral AI Support

  • Mistral Joins the Roster: The European AI powerhouse, Mistral, is now a fully supported provider, including memory embeddings and voice support.
  • Rock-solid Browser Extension: If you let your assistant control your browser, you’ll notice it drops connections way less often now. The extension’s background worker has been completely refactored to persist tab states, auto-reconnect, and strictly lock tabs during operations.

5. 🛡️ Hardcore Security Enhancements (Non-Coder Summary)

Almost half of the changelog is dedicated to absolute paranoia-level security patches.

  • Exec Approvals: Blocked sneaky ways hackers might try to bypass shell command approvals using wrappers or newline tricks.
  • File System Guards: Prevented the assistant from accidentally reading outside of its sandboxed folders (blocking symlink escapes).
  • Network Protection (SSRF): The system now strictly blocks internal/private IP ranges from being fetched unless explicitly allowed. In short: The chances of a hijacked bot compromising your local host or server have been pushed down to near zero.

Wrapping Up

While flashy new features are great, an update that fundamentally solidifies the core (security, memory, and stability) is the true mark of a great open-source project. I can’t wait to test out how sharp the new Korean memory search is!

Stay safe, update to 2.22, and enjoy your smarter AI life!