OpenClaw's latest release delivers image generation via MiniMax, native xAI search integration, ACP channel binding for Discord, and plugin approval workflows. Here's what's new and what it means for your AI agent setups.
OpenClaw just shipped another significant update — and this one might be the most visually capable release yet. March 29, 2026 brings MiniMax image generation directly into OpenClaw's agent toolkit, a native xAI search integration, ACP channel binding for Discord, and new plugin approval hooks. Let's break down each feature and how it changes the game.
The headline feature is MiniMax image-01 — a dedicated image generation model now available as part of OpenClaw's bundled providers. This isn't a plugin or external service. It's built into the core model lineup.
image-01 supports both text-to-image and image-to-image editing with aspect ratio control. Need a 16:9 hero image for an article? A 1:1 square for social? image-01 handles both with the same model. The aspect ratio control means you generate exactly the dimensions you need without cropping or upscaling artifacts.
For AI agent workflows, this opens up a new dimension. Agents can now generate visuals on demand — article thumbnails, diagram-style illustrations, feature graphics — without calling an external image API. The entire pipeline stays within OpenClaw's ecosystem.
OpenClaw now includes xAI's x_search as a first-class search tool. This is a direct integration with xAI's search capabilities, not a wrapper around an existing web search API.
What this means practically: OpenClaw agents can now search the web using xAI's infrastructure. The x_search tool is automatically enabled for any agent configured with the xAI provider. If you're running an agent with xAI as the backend, x_search is available out of the box.
For teams using OpenClaw for research tasks, this is a meaningful upgrade. Instead of configuring a third-party search tool, x_search is part of the bundled provider setup.
Perhaps the most workflow-relevant addition: ACP channel binding. The command /acp spawn codex --bind here now locks a Discord channel to a specific ACP session. Messages in that channel route directly to the bound agent — no thread management, no manual context switching.
This is a major quality-of-life improvement for teams running multiple agents. Instead of one agent handling everything in a single channel, different channels can be bound to different agents. #research routes to Jill, #coding routes to Steve, #content routes to Chad. Each agent operates in their designated lane.
The critical note: Bob (the coordinator) should NOT be bound to a channel. Coordinators need cross-channel visibility. But specialist agents — the ones focused on a single domain — benefit enormously from channel binding.
Security-conscious deployments get a new tool: before_tool_call hooks. This allows administrators to intercept and approve or reject tool calls before they execute. Combined with the new /approve command, this creates a human-in-the-loop workflow for sensitive operations.
Before a tool runs — a file write, an external API call, a system command — the hook fires. A human can review and approve or reject. Only approved calls proceed. This is essential for environments where AI agents operate with elevated permissions and zero tolerance for unintended actions.
ACP (Agent Communication Protocol) now officially supports multiple chat platforms: Discord, BlueBubbles, and iMessage. This isn't just Discord anymore. The ACP framework abstracts the chat layer, so an agent running on Discord can theoretically communicate with the same agent session via iMessage or BlueBubbles.
For personal AI setups, this means OpenClaw agents can now reach you on whichever messaging platform you prefer. For teams, it opens multi-platform support without rebuilding workflows.
Several smaller but notable changes shipped in this release:
Three themes emerge from this release: capability expansion, workflow precision, and security hardening.
The image generation alone transforms OpenClaw from a text-only agent system into a multimedia workstation. Agents that once could only describe visuals can now create them. For content teams, this eliminates a entire category of manual work.
The ACP channel binding is the workflow upgrade teams have been asking for. It solves the "which agent handles this" problem at the platform level rather than requiring custom routing logic.
And the before_tool_call hooks signal that OpenClaw is maturing for enterprise deployment. Security controls that once required custom implementation are now built in.
See also our coverage of the previous OpenClaw release for the full picture of how far this platform has come in just two days.