On March 19, ModelBest formally unveiled its new lobster-inspired AI {hardware} product, EdgeClaw Field, emphasizing safety, ease of use, and out-of-the-box performance. In collaboration with Tsinghua College and the OpenBMB group, the corporate additionally open-sourced an upgraded, safe, and environment friendly model of EdgeClaw, supporting edge-cloud collaboration to guard OPC (Operational Expertise) customers and data-sensitive enterprises amid the AI agent growth.
A key problem in deploying “lobster” AI merchandise has been excessive technical limitations, extreme Token consumption, and knowledge safety considerations—the latter being the primary impediment for OPC customers when adopting mainstream Claw-class AI instruments. EdgeClaw Field addresses these points by prioritizing safety, embedding EdgeClaw, a number one AI agent execution framework able to working each fashions and lobster brokers regionally.
Architecturally, EdgeClaw improves upon OpenClaw, resolving its limitations in native deployment and knowledge safety. On the mannequin facet, it helps mainstream cloud fashions whereas integrating ModelBest’s edge-side MiniCPM “Little Cannon” sequence. This permits easy duties to run offline, absolutely regionally, with zero Token consumption, even with out web connectivity.
A standout innovation is EdgeClaw’s self-developed Privateness Routing Middleware. By inserting hooks into the OpenClaw execution pipeline, EdgeClaw robotically categorizes person messages, software parameters, and agent outputs into three safety ranges:
- S1: Default mode – Knowledge processed within the cloud utilizing commonplace fashions.
- S2: Desensitized mode – Delicate knowledge is robotically masked earlier than cloud processing.
- S3: Safe mode – Knowledge is absolutely processed regionally, bodily remoted, utilizing pre-installed MiniCPM fashions.
EdgeClaw additionally maintains a dual-track reminiscence system: cloud fashions solely entry desensitized dialog historical past, whereas native fashions retain full knowledge, stopping privateness leaks via context home windows.
EdgeClaw preserves the mature OpenClaw structure, together with enter layers (WeChat, Feishu, Telegram), scheduling layers (model-agnostic LLM mind), execution layers (browser operations, file dealing with, API calls), and the Abilities ecosystem. It provides automated safety classification, an built-in MiniCPM LLM, and tight integration with EdgeClaw Field {hardware} for a really out-of-the-box, safe expertise.
At the moment, EdgeClaw Field is deeply appropriate with main {hardware}, together with ModelBest’s Pinea Pi, NVIDIA DGX Spark, and Apple Mac Mini, requiring no person set up or setup. Even non-technical OPC customers can deploy lobster brokers instantly.
In essence, if OpenClaw is the engine, EdgeClaw Field is the fully-built car designed for the OPC sector, that includes 4 key benefits:
- A number of – Helps a number of units, fashions, and Abilities, together with each general-purpose and industry-specific Abilities.
- Quick – Plug-and-play, able to function instantly.
- Safe – Knowledge stays native; business secrets and techniques are protected.
- Environment friendly – Easy duties run on native fashions with out Token use; complicated duties can leverage cloud fashions.
EdgeClaw Field integrates {hardware}, adaptable software program, and industry-specific Abilities—primarily forming a “digital firm.” ModelBest has dedicated to open-sourcing EdgeClaw, envisioning that in 2026, everybody ought to have their very own “lobster” AI—nevertheless it ought to run regionally, underneath the person’s management, not locked in giant tech servers. Future updates will help numerous {hardware} varieties, additional extending the imaginative and prescient of “intelligence encompassing all issues.”
Supply: IPOzaozhidao
Elevate your perspective with NextTech Information, the place innovation meets perception.
Uncover the newest breakthroughs, get unique updates, and join with a world community of future-focused thinkers.
Unlock tomorrow’s traits as we speak: learn extra, subscribe to our publication, and grow to be a part of the NextTech group at NextTech-news.com
