OpenClaw Adds xAI Multimodal Tools and Local TUI for Secure Debugging

OpenClawOpenClaw

· Updated

OpenClaw v2026.4.22 introduces full xAI multimodal support, a local terminal interface, and dynamic model registration via chat. By adding local debugging modes and token compaction tools, the update shifts the platform toward more cost-efficient and resilient self-hosted agentic workflows.

OpenClaw, an open-source AI assistant that connects models to messaging platforms, released version 2026.4.22 with full xAI integration. This update adds image generation, text-to-speech, and realtime transcription (converting audio to text instantly) for voice calls. It also introduces Tencent Hy3 and a local terminal interface for debugging.
xAI Multimodal Support
Image generation, TTS, STT, Realtime transcription
Claude Opus 4.7 Context
1M tokens
New Model Providers
Tencent Hy3, xAI
Local TUI Mode
Terminal chat without Gateway, enforced plugin gates
Token Management
Tokenjuice plugin for tool result compaction
Linux Memory Protection
OOM score adjustment for child processes

This release addresses the token tax of autonomous agents by bundling Tokenjuice, a plugin that compacts verbose tool outputs. It follows the normalized reasoning and context logic introduced in recent versions. Dynamic model registration allows users to swap providers from chat without restarting the gateway, reducing downtime during experiments.

You can now use /models add to register providers or run the local terminal interface to test plugins without a gateway. The update builds on hardened agent security and dependency repairs from the previous release. Linux users benefit from memory protections that prevent the kernel from killing the process.

OpenClaw🦞
OpenClaw🦞
@openclaw
X

OpenClaw 2026.4.22 🦞 🧠 Tencent Hy3 joins the model list 🖼️ Grok image + voice tools 🧰 Local TUI + /models add 📦 Auto-install plugins + diagnostics export Big release, tiny release notes... kidding. https://t.co/RpyRLtaNlv

85retweets870likes
View on X

Still wondering? A few quick answers below.

The Local TUI mode allows users to run terminal-based chats directly without starting the full OpenClaw Gateway. This mode is designed for quick local sessions and offline debugging while still enforcing plugin approval gates, ensuring that autonomous tools cannot execute sensitive actions without explicit user permission during the terminal session.

Tokenjuice is an opt-in plugin that reduces the number of tokens consumed during agentic workflows. It works by compacting the results of noisy tool calls, specifically those involving terminal execution and bash commands. By shrinking these outputs before they are sent back to the model, it helps lower inference costs and prevents context window overflow.

This update introduces full support for xAI multimodal tools, including image generation via grok-imagine and text-to-speech with six distinct live voices. It also adds audio transcription and realtime transcription for voice calls. These tools allow the assistant to generate visuals, speak in various formats like MP3 or WAV, and process live audio streams.

Users can now register new models directly from the chat interface using the /models add command followed by the provider and model ID. This feature allows the assistant to use the new model immediately without requiring a full restart of the OpenClaw Gateway, simplifying the process of testing different large language models or switching providers.

The OpenShell sandbox received critical hardening to prevent symlink attacks that could allow an agent to read files outside the intended directory. By pinning file reads to verified descriptors and checking the identity of parent directories, the system ensures that malicious symlink swaps cannot redirect sandbox operations to sensitive host files on the underlying machine.

Every HeadsUpAI update is written based on its original source and reviewed before it's published. Read our editorial standards →

Share this update