OpenClaw 2026.3.28: xAI Responses Integration, MiniMax Image-01, and Advanced Approval Hooks

The OpenClaw 2026.3.28 update has officially landed, continuing the momentum from the v2026.3.24 "Quality + Safety" release. This version marks a significant shift toward deep integration with xAI's real-time search capabilities and introduces a more robust security protocol for autonomous tools.

Following the roadmap established by creator Peter Steinberger (now at OpenAI), v2026.3.28 bridges the gap between raw localized intelligence and real-time world knowledge.

Real-Time Search with xAI Responses API

The headline feature of this release is the migration of the bundled xAI provider to the Responses API. This move unlocks several powerful capabilities for users leveraging Grok-powered agents.

Native x_search: No longer just a plugin, x_search is now a first-class, server-side tool. This allows agents to perform high-speed, authenticated searches across X (formerly Twitter) with superior context awareness compared to client-side scrapers.
Auto-Plugin Activation: The xAI plugin now auto-configures based on your existing web-search and tool settings. If you have an xAI key, search flows "just work" without manual toggles.
Onboarding Enhancements: New users can now set up x_search during initial onboarding, complete with a model picker and shared key management.

MiniMax Image-01: Pro-Grade Visuals

OpenClaw expands its multi-modal reach with native support for the MiniMax image-01 model. This integration provides a sleek, high-performance image generation engine directly within the OpenClaw interface.

Generation & Editing: Support for full text-to-image generation and image-to-image editing.
Granular Control: Native aspect ratio controls and prompt refinement.
Model Catalog Cleanup: To improve stability, the model catalog has been trimmed to focus on M2.7, removing deprecated legacy models (M2.1, M2.5).

The "Require Approval" Revolution

Building on our Security Best Practices, v2026.3.28 introduces the async requireApproval hook within the before_tool_call lifecycle.

This feature allows third-party tools and plugins to pause execution when a sensitive or costly action is about to occur, prompting user intervention via:

Execution Approval Overlay: A dedicated UI element in the web dashboard.
External Channels: Interactive buttons on Telegram, Discord, or BlueBubbles.
Command Line: The new /approve command, which handles both plugin and system execution approvals.

This enhancement ensures that even the most autonomous agents remain under human oversight, especially when interacting with your local file system or financial protocols like x402.

Breaking Changes & Technical Debt

As with any major update, some legacy features have been retired to make room for modernization:

Qwen Migration: The qwen-portal-auth (OAuth) integration for portal.qwen.ai has been removed. Users are encouraged to migrate to Model Studio using the provided onboarding command.
Config Cleanup: Automatic migrations for configurations older than two months have been dropped. If you are updating from a very old version, we recommend a fresh openclaw onboard to ensure your openclaw.json follows the latest schema.
JSON Schema CLI: You can now run openclaw schema --json to print the current configuration requirements, making it easier to build custom integrations.

Ecosystem & Channel Updates

The Agent Control Protocol (ACP) continues to expand its reach. OpenClaw 2026.3.28 adds current-conversation ACP binds for:

Discord
BlueBubbles / iMessage
Slack (enhanced reply logic)

Additionally, apply_patch is now enabled by default for OpenAI and OpenAI Codex models, significantly speeding up code modification tasks for developers.

OpenClaw 2026.3.28 is available now. For the full technical changelog, visit the Official OpenClaw GitHub.

New to OpenClaw? Check out our Integration Setup Guide or see how this release stacks up against the competition in our ZeroClaw vs. OpenClaw Showdown.