Yesterday, I saw that OpenClaw released an interesting update—version v2026.4.12 brings very practical improvements for people who work with AI agents. The highlights come down to two things that greatly simplify the workflow.

First, integration with LM Studio is now much more direct. Previously, you had to manually configure the address and the model name via an OpenAI-compatible interface—quite annoying. Now it has become an integrated provider that automatically detects the models you have loaded, with memory search based on local vectors—basically ready to use after the initial configuration.

Second, for those who use Feishu, a lot has changed. Bot creation is now done in a single step via QR code scanning. Before, it was a time-consuming process: you had to manually create an app in Feishu’s Open Platform, configure the callback address, and more. With Feishu integrated in this new way, the entry barrier for companies has dropped significantly.

In addition, there were improvements to the execution time of GPT-5/5.4, separating prompt construction from retry logic so that longer tasks are more stable. They also restructured how plugins are loaded—now only the components declared in the manifest are activated, making everything lighter. The Active Memory introduced in the previous version received improved search, with memory recovery changing by default to search mode, increasing accuracy across channels.

They fixed three security vulnerabilities identified with the help of AI: an empty list of approvers bypassing authorization checks, busybox/toybox being used as an interpreter for escape, and insufficient detection of shell wrappers. Solid updates for those who work with this kind of tool.