OpenClaw Update 4.12 is IMPROVING!

Summary

OpenClaw 4.12 brings significant improvements that make it more competitive with other AI coding assistants. The standout feature is the Active Memory Plugin, which works like an executive assistant by automatically searching and injecting relevant context into conversations without you needing to explicitly say "remember this." Instead of relying on keywords to trigger memory searches, the plugin proactively pulls relevant information before each response. For example, if you previously mentioned switching to a keto diet, it will automatically factor that into lunch recommendations without prompting. You can type /verbose to see exactly what context OpenClaw retrieved. The update adds two new model providers: Codex integration for OpenAI GPT models with native threading (positioning OpenClaw as an alternative to Claude Code for GPT users), and improved LM Studio support for running local models like Gemma 4 and Qwen 3.9 on your machine for free. The plugin system received a major cleanup so plugins only load what they declare, reducing context overload and improving performance. Three critical security patches address shell escape vectors by removing BusyBox and ToyBox from trusted bins. Platform fixes include resolved media fallback issues. The BoxminingAI team previously migrated to Hermes for its self-evolving skills, but these OpenClaw updates—particularly the dreaming feature and active memory—are compelling enough to keep them interested in potentially switching back. The rapid release pace (this is the second update in two days) shows OpenClaw is working hard to stay competitive in the AI coding assistant landscape.