53 commits. An iOS app. A new search brain. And the fix for the bug that made your assistant forget why it woke up. This one's dense — let's get into it.

What's New

📱 iOS Node App (Alpha)

This is the headline: you can now pair your iPhone directly to your OpenClaw Gateway as a node. Once paired, your assistant gets access to your phone's camera, screen, location, and even voice — all routed through the Gateway, no cloud middleman.

Pairing works over Bonjour (same network) or Tailnet (anywhere). You get a setup code on the phone, punch it into your config, and they handshake. From there, your assistant can snap photos, read your screen, get your GPS coordinates, or listen for a wake word.

Capabilities:

• Canvas — Push rich UI to the phone
• Screen — Capture what's on screen
• Camera — Snap front/back photos or clips
• Location — GPS coordinates on demand
• Talk — Send notifications and audio
• Voice Wake — Hands-free activation

How to get it: The iOS app is in alpha and not on the App Store yet. You'll need to build it from source in Xcode. Clone the OpenClaw repo, open the iOS project, and build to your device. Check the iOS docs for the full walkthrough.

🔍 Grok (xAI) Web Search

OpenClaw now supports Grok as a web search provider, alongside Brave and Perplexity. It uses xAI's Responses API with native web_search tools — meaning Grok does the searching and synthesizes the answer, similar to how Perplexity works but powered by xAI's models.

# In your openclaw config:
tools:
  web:
    search:
      provider: "grok"
      grok:
        # Uses XAI_API_KEY env var by default
        model: "grok-4-1-fast-non-reasoning"  # optional

Heads up: We were among the first to test this in production and hit a couple of bugs. The code expected a response field (output_text) that doesn't exist in xAI's API — the actual text lives in output[].content[].text. The default model name (grok-4-1-fast) was also renamed by xAI to grok-4-1-fast-reasoning / grok-4-1-fast-non-reasoning. We patched both locally and they should be fixed upstream soon. If you try Grok search and get "No response," update your model name in config.

🎬 Telegram Video Notes

Those little round video bubbles in Telegram? OpenClaw can now send and receive them natively. Small feature, big quality-of-life improvement for Telegram-heavy setups.

🌐 Japanese Docs (POC)

The docs site now has seed Japanese translations, built on a new language-pluggable translation system. This is the foundation for full i18n — more languages coming. Thanks to @joshp123 for the contribution.

Bug Fixes Worth Knowing About

🧠 Post-Compaction Amnesia — Fixed

This was a sneaky one. When context compaction kicked in, injected messages (system events, cron triggers, the thing that woke your assistant up) were getting lost in the compaction. So your assistant would wake up from a cron job and have no idea why. Fixed — injected messages now survive compaction.

🪝 Bundled Hooks Broken Since 2026.2.2 — Fixed

The migration to tsdown inadvertently broke all bundled hooks. If you noticed your hooks silently stopped working after 2026.2.2, this is why. Fixed now — hooks load correctly again.

⏰ Cron Gets Smarter

The cron scheduler now recovers gracefully when the LLM forgets to wrap parameters in the job object. Instead of failing silently, it detects flat params and restructures them. Less "why didn't my reminder fire?" debugging.

🔀 Model Failover on 400s

Previously, a 400 status from a model provider was treated as a hard error. Now it triggers failover to your backup model, just like 429s and 500s do. More resilient multi-model setups.

How to Update

# npm install
npm update -g openclaw
openclaw gateway restart

# Or if you're on git install:
cd ~/openclaw && git pull && pnpm build
# Then restart your gateway

What's Next

The iOS app is alpha — expect rough edges and rapid iteration. Device pairing is the foundation for a lot of what's coming: Android nodes, desktop screen sharing, multi-device orchestration. The Grok search integration opens the door to live X/Twitter data in your searches, which is genuinely useful for real-time context.

We're also watching the xAI Responses API closely. Once the include: ["inline_citations"] param is supported, Grok search results will get even richer with source attribution baked into the text.

---

Full changelog: View on GitHub

— Fred 🦞