Release Notes

8 min read

OpenClaw 2026.4.27: Codex Computer Use & DeepInfra Integration

Bundled DeepInfra support, Codex Computer Use desktop control, expanded channel coverage, and GPU-enabled sandboxing.

Release Overview

OpenClaw 2026.4.27 focuses on Desktop Control and Provider Diversity. With the official introduction of Codex Computer Use and DeepInfra integration, agents are becoming more capable of interacting with both the physical desktop environment and specialized high-performance model providers.

Major Technical Breakthroughs

Codex Computer Use

Desktop automation reaches a new level of reliability. OpenClaw now includes a dedicated setup for Codex Computer Use.

  • Discovery: Marketplace discovery for cua-drivers and MCP servers.
  • Reliability: Fail-closed MCP checks ensure that your desktop control environment is ready before an agent starts its turn.
  • CLI Tools: New /codex computer-use status and install commands for rapid setup.

DeepInfra Bundled Support

DeepInfra joins the core provider set, offering a wide range of capabilities:

  • Media Suite: Native media generation and editing support.
  • AI Toolkit: Integrated TTS, embeddings, and automatic model discovery.
  • Performance: Highly optimized inference for popular open-weights models.

Expanded Channel Coverage

Connecting OpenClaw to your preferred communication platforms continues to expand.

  • Tencent Yuanbao: New external channel plugin support with a dedicated quick-start guide.
  • QQBot: Full group chat support, including history tracking, @-mention gating, and large file uploads via chunked delivery.

Core System Improvements

GPU Passthrough for Sandboxes

Performance-heavy tasks like local image processing or model fine-tuning can now leverage hardware acceleration. The new sandbox.docker.gpus option allows sandboxed agents to access host GPUs via the --gpus Docker flag.

Manifest-First Metadata

Gateway startup is now faster and more auditable. Model catalogs and plugin startup now use manifest-first metadata. This reduces boot time and makes it easier to audit provider aliases and suppressions.

Presence & Security

  • Node Presence: New protocol events for iOS and Android allow paired nodes to be marked as "recently alive" without requiring a constant active connection.
  • Outbound Proxies: Added support for operator-managed outbound proxy routing with strict validation for enterprise environments.

Upgrade Guide

Who should upgrade?

  • Users implementing Codex desktop control (Computer Use)
  • Operators needing DeepInfra for media or high-speed inference
  • Developers running GPU workloads in sandboxed containers
  • Users on QQBot or Tencent Yuanbao channels

How to Upgrade

# Perform the update
openclaw update

# Check Codex Computer Use status
/codex computer-use status

For the complete list of fixes and technical changes, visit the official release page on GitHub.

FAQ

What is Codex Computer Use in OpenClaw?

It is a new setup for Codex-mode agents that allows desktop control. It includes status/install commands, marketplace discovery for cua-drivers, and fail-closed MCP checks to ensure safety before turns start.

How do I use the new DeepInfra provider?

DeepInfra is now a bundled provider. It supports automatic model discovery, media generation/editing, TTS, and embeddings. You can enable it in your provider settings with your DeepInfra API key.

What are the improvements for QQBot and Yuanbao?

Tencent Yuanbao now has a dedicated external channel plugin and quick-start guide. QQBot support has been expanded to include full group chat functionality, history tracking, and high-performance streaming message delivery.

How do I enable GPU support for sandboxed agents?

You can now use the opt-in 'sandbox.docker.gpus' setting. When enabled, it passes the --gpus flag to your Docker sandbox containers, allowing local GPU workloads to run inside sandboxed agents.

Need help from people who already use this stuff?

Command Your Desktop

Ready to try the new Codex Computer Use features? Join the community to share your desktop automation scripts and learn how to optimize your DeepInfra workflows.