OpenClaw v2026.4.22: xAI Takes Over, TUI Goes Local, Tencent Joins

OpenClaw shipped v2026.4.22 today — one of the most feature-dense releases in recent memory. Here’s what’s landing:

xAI is now a first-class citizen. Grok image generation (grok-imagine-image, grok-imagine-image-pro), reference-image editing, six TTS voices, MP3/WAV/PCM/G.711 output formats, speech-to-text transcription, and realtime transcription for Voice Call streaming — all via the xAI provider.

STT gets streaming too. Deepgram, ElevenLabs, and Mistral now support Voice Call streaming transcription alongside the existing OpenAI and xAI paths. ElevenLabs also gains Scribe v2 batch audio transcription.

TUI goes local. A new local embedded mode lets you run terminal chats without a Gateway connection, while still enforcing plugin approval gates.

Onboarding finally works out of the box. Missing provider and channel plugins now auto-install during setup — no more manual recovery on first run.

WhatsApp gets smarter. Configurable native reply quoting (replyToMode), and per-group/per-direct systemPrompt forwarding via GroupSystemPrompt with wildcard (*) support and account-scoped overrides.

New provider: Tencent Cloud. Bundled TokenHub onboarding, docs, and Hy3-preview model catalog entries with tiered pricing metadata.

GPT-5 prompt overlay goes universal. The GPT-5 behavior overlay is now shared across OpenAI, OpenRouter, OpenCode, Codex, and all other GPT providers — not just the OpenAI plugin. A new agents.defaults.promptOverlays.gpt5.personality toggle controls it globally.

Other notable changes:

  • Amazon Bedrock Mantle now supports Claude Opus 4.7 via bearer-auth streaming
  • OpenAI Codex auth import removed from onboarding — use browser login or device pairing instead
  • /models add command added so you can register a model from chat without restarting the gateway
  • Diagnostics export now includes sanitized logs, config, health, and stability snapshots by default
  • OpenAI-compatible backends (vLLM, SGLang, llama.cpp, LM Studio, LocalAI, Jan, etc.) now correctly track token usage
  • ACPX adds an openClawToolsMcpBridge option injecting a core OpenClaw MCP server for built-in tools (starting with cron)
  • CLI/doctor plugin lazy-loading cut cold startup time by ~74%

The full changelog is on GitHub.

← Back to News