🦞 OpenClaw 2026.2.24

What's New + What Matters

February 25, 2026 · Boxmining

215K+

GitHub Stars

Total Releases

373K

X Followers

📢 The Official Drop

2,173 ❤️ · 179K views · 176 RTs — Feb 25, 2026

@openclaw (373K followers)

OpenClaw 2026.2.24 🦞

🌍 Stop phrases in 10+ languages (your bot finally understands "arrête")
⌨️ Typing indicators that don't ghost you
🪟 PowerShell 7 because it's not 2019
🔒 30+ security fixes (we don't sleep so you can)

Updating is self-care.

2,173 ❤️ · 176 🔁 · 62 quotes · 179K views · view tweet

🆕 What's New in v2026.2.24

🌍

Multilingual Stop Phrases

"stop", "arrête", "para", "停止", "やめて", "रुको" — your agent now understands abort commands in 10+ languages including ES/FR/ZH/HI/AR/JP/DE/PT/RU. Also accepts "STOP OPENCLAW!!!" with trailing punctuation.

📱

Android App Overhaul

Native 4-step onboarding → 5-tab shell (Connect, Chat, Voice, Screen, Settings). Full Connect setup/manual mode. Refreshed chat + settings UI.

⌨️

Typing Indicators Fixed

Keepalive refresh during long replies. No more typing indicator disappearing mid-inference. Works across Discord, Telegram, WhatsApp.

🔒

30+ Security Fixes

Docker namespace-join blocked by default. Multi-user trust model heuristic. Heartbeat DM leakage prevented. WhatsApp reasoning block suppression.

🛡️ Security: The Big Ones

30+ fixes — they're not messing around

Docker namespace-join blocked — container:* mode disabled by default for sandbox. Break-glass opt-in only.
Multi-user trust model — new heuristic flags shared-user ingress. Hardening guidance for multi-user setups (sandbox.mode="all", workspace-scoped FS).
Heartbeat DM leakage fixed — heartbeats no longer spam your Discord DMs or Telegram. Delivery blocks direct-chat targets.
WhatsApp reasoning suppression — hidden thinking blocks no longer leak to end users. Hard-drops "Reasoning:" prefixed text.
Routing isolation hardened — cross-channel replies can't hijack Discord-targeted messages in shared sessions.

Combined with v2026.2.23's 50 advisories + 6 CVE patches, OpenClaw has shipped ~80 security fixes in 2 days. Enterprise-grade hardening.

🔧 Platform Fixes That Matter

🎙️

Discord Voice Reliability

DAVE encryption restored, configurable join options, auto-rejoin after decrypt failures. STT finally stable.

📱

WhatsApp Reconnect

Status 440 handled properly — no more infinite reconnect loops. Clean relink guidance on session conflicts.

🤖

Model Fallback Chain

When primary model is in cooldown, fallback chain now traverses properly instead of dead-ending. No more stuck agents.

🇨🇳

CJK Input Fix (macOS)

Return key now confirms IME candidates first — no more accidental sends while composing Chinese/Japanese/Korean text.

🗣️ Talk Mode Gets Real

Provider-agnostic voice config + ElevenLabs integration

Provider-agnostic Talk config — no longer locked to one TTS provider. Legacy configs auto-migrate.
ElevenLabs metadata — gateway exposes setup/status for voice configuration surfaces.
macOS mic guard — all audio input paths handle missing microphones gracefully. No more crashes on mic-less Macs.
Discord DAVE encryption — configurable encryption + failure tolerance for voice channels.

Voice is getting serious attention. Still early compared to dedicated voice assistants, but the foundation is solid for real-time conversations.

⚠️ Breaking Changes

Two things that might bite you on upgrade

1. Heartbeat DM Delivery Blocked

Heartbeats no longer send to direct/DM targets (user:, Telegram user chats, WhatsApp direct numbers). Only channel/group targets receive heartbeat messages. Runs still execute — just no DM spam.

2. Docker Namespace-Join Blocked

network: "container:*" mode blocked by default for sandbox + sandbox-browser containers. To re-enable: set dangerouslyAllowContainerNamespaceJoin: true

Both are security-positive changes. Most users won't notice. Docker power users: check your compose files.

📋 Recap: v2026.2.23 (Yesterday)

PREV In case you missed it

Kilo Gateway provider — native integration, full auth stack, enterprise-ready
Moonshot/Kimi vision + video — analyze video content, not just images
Claude Opus 4.6 support — latest Anthropic model, 1M context
50 security advisories triaged → 12 survived as real issues, 6 CVEs patched
SSRF protection — defaults to trusted-network mode. Run openclaw doctor --fix to migrate.
Config snapshot redaction — API keys auto-hidden in logs, diagnostics, OTEL telemetry

🎯 February's Game Changers

CONTEXT 9 releases in 25 days

🧠

Multi-Model Freedom

15+ providers: Claude, GPT-5.3, Gemini, Grok, Mistral, Moonshot, Ollama local. Switch per-agent or per-session.

🤖

Sub-Agent Orchestration

Spawn parallel workers with scoped permissions. QC pipeline built in. The real unlock for complex workflows.

👁️

Vision + Video

Moonshot/Kimi: analyze video content natively. Not just screenshots — actual video understanding.

⌚

Apple Watch + iOS

Wrist notifications, share extension (forward anything to your agent), voice input from Watch.

💬 What People Are Saying

@Aipromptslap — "Finally getting PowerShell 7 and stop phrases that actually work. 🦞 High-key needed that 'arrête' fix for multilingual workflows."

link

@chriskhan01 — "Reading OpenClaw release notes feels like watching the category mature: Fixing update delivery, cleaning stale tokens, and smoothing auth edge cases."

2 ❤️ · 51 views · link

@MukAshPatel — "Major update by @steipete — Android app → 4-step onboarding → 5-tab shell: Connect, Chat, Voice, Screen, Settings"

link

@_karrtopelka — "Feature idea: native Telegram topic-based routing. A group with topics where each one maps to a different agent/model"

link

⚡ The Real Talk

Not everyone's having a smooth ride

@yepsurethatsme — "Did anyone else's OpenClaw go full retard after installing yesterday's update? Mine is way slower, does way dumber stuff and is generally useless"

46 views · link

@rubio_josea — "Is it possible to downgrade? Today I've been hitting my 1M API limit with no apparent reason"

link

@realsoulciti — "Caught my bot making stuff up. When I pointed out its fabrication it tried to get me off the issue."

link

Some users report performance regression after update. Token consumption spikes reported. The $100 learning curve is still real.

⚔️ Meanwhile: Claude Code Fires Back

@padawanjoy — "OpenClaw 🦞 might need to watch its back. Anthropic just dropped a major update for Claude Code: Official Remote Control support"

3 ❤️ · link

Dimension	OpenClaw 2.24	Claude Code	Cursor
What it is	Life automation	Coding agent	AI IDE
Interface	50+ channels	Terminal/IDE	VS Code fork
Models	15+ providers	Claude only	Multiple
Self-hosted	Fully local	No	No
Coding	Basic	Superior	Superior
Autonomy	24/7 + cron	Session-based	Session-based
Voice	Talk + ElevenLabs	No	No
Mobile	Android+iOS+Watch	No	No

Different tools, different jobs. OpenClaw = life OS. Claude Code = coding beast. Power users run both.

✅ What's Actually Useful in 2.24

✅ Multilingual Stop Phrases

Run agents in non-English contexts? This is huge. Finally abort in your native language.

✅ Heartbeat DM Fix

No more random heartbeat spam in your DMs. Should've been fixed ages ago.

✅ Android Redesign

Proper onboarding + 5-tab nav. Mobile setup actually doable for new users.

⚠️ Model Fallback Fix

Agents no longer stuck when primary model is down. But configure fallbacks first.

Stability + polish release. No flashy features — but fixes that make daily use less frustrating.

🔮 The Bigger Picture

9 releases in 25 days — where is this going?

Security-first pivot — 80+ security fixes in Feb alone. Enterprise customers pushing this.
Mobile is real — Android overhaul + Apple Watch + iOS share extension. Not an afterthought.
Voice is next — provider-agnostic Talk + ElevenLabs + Discord DAVE. Always-on voice assistant.
Multi-user hardening — trust model heuristics, workspace scoping. Teams use case incoming.
215K GitHub stars — Wikipedia page, 373K X followers. Category leader.

🦞 Should You Update?

Yes

If you use Discord voice, WhatsApp, or non-English languages

Wait

If current setup works — some users report regressions

How to update:
npm update -g openclaw && openclaw gateway restart

If things break:
openclaw doctor --fix

🦞

OpenClaw 2026.2.24

Stability release. 30+ security fixes. Multilingual. Android redesign.

The lobster doesn't sleep. Neither should your agent.

GitHub Release · Official Tweet · Docs