🦞 OpenClaw 3.12 Just Dropped
Good Morning, OpenClaw Owners!
Agents are beginning to grow its own nervous system.
OpenClaw 3.12 Just Dropped — MyClaw.ai Already Updated
OpenClaw just released v2026.3.12, and MyClaw.ai is already running it.
If you’re hosting your agents on MyClaw, you can upgrade your instance to 3.12 immediately and get the newest improvements without touching installs, configs, or deployments.
Highlights from this release:
- 🖥️ Refreshed Control UI — modular dashboard, command palette, mobile tabs, better chat tools
- ⚡ Fast mode toggles — configurable speed mode for OpenAI and Anthropic across CLI, TUI, Control UI, and ACP
- 🔌 Provider plugin system — Ollama, vLLM, and SGLang now run through modular provider plugins
- ☸️ Kubernetes deployment support — raw manifests, Kind setup, and full deployment docs
- 🤖 Subagent sessions_yield — orchestrators can end a turn instantly and pass payloads to the next session
- 🛡️ Security hardening — device pairing tokens, plugin restrictions, exec approval fixes, webhook signature enforcement
The OpenClaw ecosystem keeps moving fast.
Sign up on MyClaw to start directly with OpenClaw 3.12, or upgrade your existing instance to 3.12 with one click. 🚀
OpenClaw Builds Messaging Testbed
TL;DR: Peter Steinberger said he is building a full end-to-end testing bed for OpenClaw, turning messaging channels like Telegram and WhatsApp into scriptable, CLI-driven infrastructure. The goal is to verify channel behavior automatically across environments, so OpenClaw can keep shipping fast while catching regressions before broken messaging workflows reach users—a big step toward making agent messaging feel like reliable software, not fragile glue code.
OpenClaw Phone Idea
TL;DR: Midjourney founder David Holz posted that he wishes smartphones could be replaced by a pocket-sized color e-ink device running only one app: OpenClaw. In replies, he clarified the device wouldn’t run apps locally but would let the agent control a remote computer with full software.
Perplexity Faces OpenClaw Reality
TL;DR: A viral post mocked Perplexity’s trajectory—from building an AI search engine to raising a $21B valuation and launching “Perplexity Computer”—claiming it would “kill OpenClaw,” which is free. The claim drew skepticism online, with many arguing the products solve different problems: Perplexity focuses on desktop automation and interfaces, while OpenClaw is an open orchestration layer developers can run themselves.
Peter Challenges Tencent Mirror
TL;DR: OpenClaw creator Peter Steinberger criticized Tencent after discovering on X that the company had launched “SkillHub,” a platform mirroring all skills from ClawHub. Steinberger said transparency would have meant contacting the project first, not quietly launching a mirror. Tencent replied that SkillHub is a localized mirror for Chinese users and credited the original source.
OpenClaw Electric Vehicle
TL;DR: Chinese mobility company Ninebot announced support for OpenClaw integration, bringing AI agents into its two-wheel electric vehicle ecosystem. A new skill, ninebot-device-skill, has been released on ClawHub, allowing agents to read vehicle data such as battery level, charging status, location, mileage, and range estimates. Early access testing opens March 13, with current capabilities limited to read-only information.
China Restricts Official OpenClaw Use
TL;DR: Bloomberg reported that Chinese authorities have begun restricting the installation of OpenClaw on office computers used by government agencies and state-owned enterprises, including major banks, citing security risks. Some employees were told to report existing installations for review or removal. The move contrasts with local policy proposals simultaneously encouraging OpenClaw ecosystem development through subsidies and deployment support.
NVIDIA Teaches OpenClaw at GTC
TL;DR: NVIDIA announced a hands-on “Build-a-Claw” experience at its GTC conference (March 16–19) where attendees can learn to create a long-running AI agent using the open-source OpenClaw framework. Participants will configure tools, customize agent behavior, and deploy an always-on agent during the session.
Agents Fixing Agents
TL;DR: Developer Nick Vasile proposed running multiple agents inside an OpenClaw environment, suggesting users install Claude Code so one agent can repair problems created by another. The idea frames a new workflow where OpenClaw orchestrates tasks while coding agents monitor, debug, and patch failures, creating a multi-agent system where agents supervise and fix each other.
Run OpenClaw on MyClaw.ai - No setup, No tech skills, Always-on 👇
