OpenClaw v2026.4.22

Multimodal, Voice & Embedded Mode

April 23, 2026

🎯 Executive Summary

OpenClaw v2026.4.22 is a substantial feature release adding xAI multimodal support, Voice Call streaming across five STT providers, a local embedded TUI mode, auto-installing plugins, and native OpenAI web_search. It also expands provider coverage to Tencent Cloud and Amazon Bedrock Mantle.

  • xAI image generation, TTS, and STT support
  • Voice Call streaming for 5 STT providers
  • TUI embedded mode (no Gateway required)
  • Auto-install missing plugins during onboarding
  • OpenAI native web_search tool
  • Tencent Cloud & Amazon Bedrock Mantle providers

🎨 xAI Multimodal Support

  • Image generation: grok-imagine-image / grok-imagine-image-pro
  • Reference-image edits — upload an image, get variations
  • Text-to-Speech: Six live xAI voices
  • Formats: MP3, WAV, PCM, G.711
  • STT: grok-stt audio transcription
  • Realtime: xAI realtime transcription for Voice Call streaming

🎧 Voice Call Streaming Transcription

New streaming transcription providers for Voice Call:

  • Deepgram — realtime streaming
  • ElevenLabs — realtime + Scribe v2 batch transcription
  • Mistral — realtime streaming
  • OpenAI — existing path, hardened
  • xAI — new realtime path

Shared WebSocket transport and batch form helpers reduce provider boilerplate.

💼 TUI Embedded Mode

Run terminal chats locally without a Gateway while keeping plugin approval gates enforced. This makes OpenClaw usable in air-gapped or lightweight environments where running a full Gateway is undesirable.

Key benefit: local execution with remote-grade safety.

🔧 Developer Experience

  • Auto-install missing plugins during onboarding — no manual recovery
  • /models add <provider> <modelId> — register models from chat without restart
  • OpenAI native web_search — automatic for direct Responses models
  • WhatsApp native reply quoting with replyToMode
  • Per-chat system prompts for WhatsApp groups and DMs

☁️ New Cloud Providers

Tencent Cloud

  • TokenHub onboarding
  • hy3-preview model catalog
  • Tiered Hy3 pricing metadata

Amazon Bedrock Mantle

  • Claude Opus 4.7 via Mantle's Anthropic Messages route
  • Provider-owned bearer-auth streaming
  • No need to treat AWS bearer tokens like Anthropic API keys

🛠️ Reliability & Performance

  • Plugin load time reduced 82-90% via native Jiti loading
  • doctor --non-interactive runtime reduced ~74% via lazy loading
  • Linux OOM protection for child processes via oom_score_adj
  • Config reload without restart for 10+ plugins
  • Gateway restart preserves chat context and one-shot continuations

🔐 Security Hardening

  • Auth/commands: require owner identity for owner-enforced commands
  • Dotenv sandboxing for Matrix, Mattermost, IRC, Synology endpoints
  • Slack thread alias preservation prevents cross-thread leaks
  • Browser rejects invalid accessibility refs immediately
  • Plugin integrity drift detection on update

🌐 Deployment

Release: github.com/openclaw/openclaw/releases/tag/v2026.4.22

v2026.4.21 (previous day): GPT-Image-2 default, auth hardening, plugin doctor fixes

Upgrade: npm install -g openclaw or openclaw update