ME News reports that on April 14 (UTC+8), the open-source AI Agent tool OpenClaw released version v2026.4.12. This update focuses on expanding model and channel integration: it adds LM Studio as a built-in provider, enabling local models to work out of the box; the Feishu channel now supports QR code-based bot creation, simplifying enterprise onboarding.
Previously, LM Studio required users to manually configure the address and model name via a generic OpenAI-compatible interface. This version upgrades it to a built-in provider that supports automatic discovery of models loaded in LM Studio, streamed preloading, and memory search based on local embedding vectors—ready to use after configuration.
Previously, the Feishu channel required manually creating an application and configuring the callback URL on the Feishu Open Platform, resulting in a lengthy process. The new version supports one-step robot creation via QR code scanning, lowering the barrier to entry for enterprise users.
Other user-facing changes:
1. Runtime improvements for GPT-5/5.4: separated prompt construction and retry logic for greater stability in long-running tasks.
2. Plugin loading refactored: Only the necessary components declared in the manifest are activated at startup; unrelated plugin runtimes are no longer loaded.
3. The newly added Active Memory plugin in the previous version improved search behavior, with memory recall defaulting to search mode and enhanced cross-channel accuracy.
Three vulnerabilities have been patched for security, all marked as AI-assisted discoveries: empty approver list bypassing authorization checks, busybox/toybox being usable as an interpreter escape, and insufficient shell wrapper detection leading to environment variable injection. (Source: GitHub)