Open-Source AI Agent Platform OpenClaw Releases v2026.3.22 With GPT-5.4 Support

OpenClaw, an open-source AI agent framework popularly known in its community as “Lobster,” connects large language models with messaging platforms, allowing automated assistants to be deployed in everyday conversations. The update matters because voice interaction and Prompt Cache cost controls directly affect how agents are used, API spending and service stability.

OpenClaw released the medium-sized v2026.5.4 update on May 4, 2026, adding a Gemini voice bridge that enables AI agents to make voice calls. It also fixed a Prompt Cache issue to prevent cache failures from causing API costs to surge, though the project did not disclose specific savings. The release also improves WhatsApp, Telegram and Discord functionality and strengthens several security safeguards.

OpenClaw is an open-source platform for deploying and managing AI agents. In industrial and other settings involving continuously running tasks, an agent that cannot accept mid-process corrections may amplify operational and cybersecurity risks. Previous incidents involving malicious ClawHub plugins also made plugin supply-chain protection a key focus of this update.

OpenClaw released stable version 2026.5.3 on May 3, 2026, adding a /steer command that lets users intervene in an agent's work in real time and redirect it while it is running. The new version also restricts file transfers for nodes that have not completed pairing, while strengthening ClawHub plugin safeguards and security patches to reduce the risk of malware exploitation.

OpenClaw is an open-source AI agent framework that coordinates models, tools and subagents to carry out complex tasks. The update integrates OpenAI’s image model and enhances collaboration among multiple layers of agents, reducing the burden of API-key management for developers and allowing lengthy workflows to be broken into smaller tasks. It also patches high-risk vulnerabilities, an important improvement for enterprise deployments and data security.

OpenClaw released v2026.4.23 on April 23, 2026, adding the ability to sign in through Codex OAuth and call OpenAI’s gpt-image-2 directly without entering a separate API key. Subagents can now be nested up to three levels deep and inherit conversation context from their parent agents. The release also patches several Critical-rated security vulnerabilities, strengthening protections for agent execution environments.

OpenClaw is a system that uses AI agents to operate tools and run workflows across communication platforms. Agents can invoke plugins autonomously, but that capability also raises the risk of errors or unauthorized actions. The addition of human approval checkpoints allows high-risk plugin operations to be confirmed before an agent proceeds.

OpenClaw released version 2026.3.28 on March 28, 2026, ahead of ClawCon in Tokyo, introducing “plugin approval hooks.” The update also integrates the xAI Responses API to give Grok search capabilities and improves workspace binding for Discord and iMessage.

OpenClaw is an open-source AI coding agent designed for lightweight deployment and collaboration across agents. As tools including Anthropic’s Claude Code and OpenAI Codex CLI compete for a place in developers’ workflows, integration with different agents and communication standards has become critical for open-source platforms seeking to expand their ecosystems and strengthen their competitiveness.

OpenClaw announced in July 2026 that an update scheduled for the following week would add support for Claude Code and OpenAI Codex CLI plugins. The new version will separate the core engine and introduce the ACP protocol to improve cross-agent communication. Founder Peter Steinberger remains personally involved in writing code. The announcement did not include any funding amount or pricing plans.