OpenClaw 2026.4.22: xAI Image Generation, TUI Embedded Mode & /models Add

Headline Features

xAI Image Generation & Voice

The xAI provider now supports comprehensive media capabilities. You can generate images with grok-imagine-image and grok-imagine-image-pro, including reference-image edits. For voice, there are six live xAI voices, multiple TTS formats (MP3/WAV/PCM/G.711), and grok-stt audio transcription. Realtime transcription for Voice Call streaming is also supported.

// xAI image generation
{
  "image": {
    "generate": {
      "prompt": "A futuristic cityscape at sunset",
      "provider": "xai"
    }
  }
}

// xAI text-to-speech
{
  "tts": {
    "text": "Hello from OpenClaw",
    "voice": "coral",
    "format": "mp3"
  }
}

TUI Embedded Mode

The Terminal UI now supports local embedded mode for running terminal chats without a Gateway while keeping plugin approval gates enforced. This is perfect for local development, offline work, or when you want a lightweight setup without the full Gateway infrastructure.

/models Add Command

You can now register models dynamically from chat without restarting the gateway. The new /models add command accepts provider and model ID, making it easy to try new models as soon as they are available. The existing /models command remains a simple provider browser with clearer add guidance and copy-friendly command examples.

# Add a new model from chat
/models add openai gpt-5
/models add anthropic claude-opus-4
/models add xai grok-3

# Browse available models
/models

WhatsApp Reply Quoting

WhatsApp now supports configurable native reply quoting with replyToMode for both group and direct conversations. Additionally, per-group and per-direct systemPrompt config is forwarded into inbound context, so configured per-chat behavioral instructions are injected on every turn.

Voice Call & STT Improvements

Voice Call Streaming Transcription

Realtime transcription for Voice Call streaming is now supported across multiple providers: Deepgram, ElevenLabs, Mistral, OpenAI, and xAI. ElevenLabs also gains Scribe v2 batch audio transcription for inbound media, providing more accurate and faster transcription.

STT Provider SDK

A new Plugin SDK for STT providers shares realtime transcription WebSocket transport and multipart batch transcription form helpers. This reduces provider plugin boilerplate while preserving proxy capture, reconnects, audio queueing, close flushing, upload filename normalization, and ready handshakes.

Provider Updates

Tencent Cloud Provider

A new bundled Tencent Cloud provider plugin is available with TokenHub onboarding, documentation, hy3-preview model catalog entries, and tiered Hy3 pricing metadata for users in China and surrounding regions.

Amazon Bedrock Mantle

Claude Opus 4.7 is now available through Mantle's Anthropic Messages route with provider-owned bearer-auth streaming. The model is callable without treating AWS bearer tokens like Anthropic API keys, simplifying authentication for Bedrock users.

GPT-5 Prompt Overlay

The GPT-5 prompt overlay has been moved into the shared provider runtime, so compatible GPT-5 models receive the same behavior and heartbeat guidance through OpenAI, OpenRouter, OpenCode, Codex, and other GPT providers. A new global toggle agents.defaults.promptOverlays.gpt5.personality controls the friendly-style overlay.

OpenAI-Compatible Providers

Known local backends (vLLM, SGLang, llama.cpp, LM Studio, LocalAI, Jan, TabbyAPI, text-generation-webui) are now marked as streaming-usage compatible. Token accounting no longer degrades to unknown/stale totals for these servers. Additionally, streamed token usage is recovered from llama.cpp-style timings metadata.

CLI & Developer Experience

Claude CLI Warm Sessions

Claude CLI runs now default to warm stdio sessions, including custom configs that omit transport fields. Sessions resume from the stored Claude session after Gateway restarts or idle exits, providing a smoother experience for Claude Code users.

Codex Harness Improvements

The Codex harness now routes native app-server turns through before_prompt_build, fires before_compaction / after_compaction for native compaction items, and fires llm_input, llm_output, and agent_end for lifecycle hooks. Tool call handling is improved with async tool_result middleware and after_tool_call events.

Plugin Loading Optimization

Native Jiti loading is now preferred for built bundled plugin dist modules on supported runtimes, cutting measured bundled plugin load time by 82-90%. The CLI doctor command also lazy-loads plugin paths and prefers installed plugin dist runtime entries, reducing doctor --non-interactive runtime by about 74%.

Onboarding Improvements

Missing provider and channel plugins are now auto-installed during setup, so first-run configuration can complete without manual plugin recovery. The Codex CLI auth import path has been removed from onboarding to avoid copying OAuth material into agent auth stores; use browser login or device pairing instead.

Other Notable Changes

• Agents/Sessions: Mailbox-style sessions_list filters for label, agent, and search plus visibility-scoped derived title and last-message previews
• Control UI: Browser-local personal identity for the operator (name plus avatar), improved Quick Settings and chat layouts
• Gateway Diagnostics: Payload-free stability recording by default, support-ready diagnostics export with sanitized logs and snapshots
• Pi Models: Updated to 0.68.1 with refreshed opencode-go/kimi-k2.6, Qwen, GLM, MiMo, and MiniMax entries
• Tokenjuice: Bundled native OpenClaw support for compacting noisy exec and bash tool results in Pi embedded runs
• ACPX: Explicit openClawToolsMcpBridge option injecting core OpenClaw MCP server for selected built-in tools
• Docs: Thai translation support added

How to Update

Update to 2026.4.22 with one command:

openclaw update

Or tell your agent:

Update OpenClaw to the latest version

The update typically completes in under 30 seconds. Your agents will automatically use the new features without configuration changes.