OpenClaw v2026.4.22 shipped April 22, and most coverage led with the xAI integration — Grok image generation, 6 TTS voices, real-time STT. That's the flashy headline.

But for operators running OpenClaw day-to-day as production infrastructure, the more consequential changes are the stability and usability fixes buried lower in the changelog. Duplicate messages. Session interruptions. Config loss on restart. WhatsApp group behavior. Live model management without a restart.

This post focuses on what actually changed for the daily reliability of your instance.

Update Command

npm update -g openclaw && openclaw gateway restart

Always restart the gateway after updating. The stability fixes in this release are gateway-side — they don't take effect until the process restarts.

The Stability Changes That Matter

Reliability

Duplicate Message Fixes

Several paths that caused OpenClaw to deliver the same message multiple times have been patched in this release. This was one of the more common complaints for WhatsApp and Discord users on long-running instances — the agent would respond correctly, then repeat itself after a few seconds.

If you've been seeing duplicates: update and restart. This should resolve the most common duplication paths.

WhatsApp

Native Reply Quoting + Per-Group System Prompts

Two meaningful WhatsApp improvements in this release:

Native reply quoting: OpenClaw now supports configurable native reply quoting (replyToMode for WhatsApp conversations). When your agent replies, it properly threads back to the original message — cleaner UX on mobile, especially in active group chats.

Per-group system prompts: You can now configure a different systemPrompt per WhatsApp group (or per direct chat), with "*" wildcard support. This is major for operators running one OpenClaw instance across multiple WhatsApp groups — each group can have a different persona, scope, and set of instructions. The config now forwards GroupSystemPrompt into inbound context on every turn.

Models / CLI

/models add — Register Models Live, No Restart Required

New command: /models add <provider> <modelId>

Previously, adding a new model required editing config and restarting the gateway. Now you can register a model directly from chat and use it immediately — no restart, no downtime. The model becomes available to the current session instantly.

This matters most for operators who iterate on model configuration frequently, or who want to test a new model without interrupting a running agent. Combined with the existing /model command for switching, this makes model management a live operation rather than a restart-required config change.

Onboarding / Setup Wizard

Redesigned Setup Wizard + Auto-Install Plugins

The setup wizard got a meaningful UX refresh in this release:

  • Security disclaimer now uses a single yellow warning banner with section headings and bulleted checklists — the key guidance is no longer easy to miss
  • Loading spinner during the initial model catalog load — the wizard no longer goes blank for 10–20 seconds while it fetches providers
  • "API key" placeholder on provider API key prompts — clearer what's expected
  • Missing provider and channel plugins now auto-install during setup — first-run configuration can complete without manual plugin recovery

If you've tried setting up OpenClaw before and hit the blank-screen-during-setup problem or the "plugin not found" error on first run, these changes address both.

Models / CLI

Kimi K2.6 Rows + Smarter Provider Handling

The models list --all command now shows bundled provider-owned static catalog rows (including Kimi K2.6 rows for Moonshot, OpenRouter, and Vercel AI Gateway) before auth is configured — so you can see what's available without needing to authenticate first.

models list --provider now validates that you pass provider IDs, not display labels, and gives a clear error before loading model discovery if you get this wrong. Cleaner error messages = less time debugging model config.

TUI

Local Embedded Mode (No Gateway Required)

New: --local / embedded mode for running terminal chats without a Gateway. Useful for quick local testing, scripts, or CI pipelines where you don't want to spin up the full gateway process. Plugin approval gates are still enforced in this mode.

The xAI Summary (Quick Version)

For completeness — the headline features covered elsewhere:

Should You Update Now?

Yes. The duplicate message fixes and WhatsApp per-group system prompts alone are worth it for most operators. The setup wizard changes matter if you're onboarding new users. The /models add command is a quality-of-life improvement that will compound over time.

Security reminder: While you're updating — verify your gateway is bound to 127.0.0.1, not 0.0.0.0. The recent CVE-2026-41352 and CVE-2026-41349 patches were in earlier releases, but misconfig is still the most common exposure vector. openclaw gateway status will show your bind address.

Changelog Reference

Full release notes available at newreleases.io and the openclaw/openclaw GitHub repo.

Running an Older Version?

If you're not on a recent version, you may be missing security patches from the last 3 releases. A ClawReady audit checks your version, binding config, and security posture — and documents every remediation step.

Get a Security Audit →