EdgeClaw - An open-source AI agent framework developed by Wallfacer in collaboration with Tsinghua University and other institutions.

What is EdgeClaw?

EdgeClaw is an open-source AI agent framework jointly launched by Mianbi Intelligence, Tsinghua University, and OpenBMB, focusing on edge-cloud collaboration, local deployment, and secure control. The framework pioneered a three-layer data security protocol (S1/S2/S3), enabling public information to be uploaded to the cloud, sensitive data to be anonymized, and private content to be locked locally. Combined with a dual-track memory mechanism and the GuardAgent protocol, it achieves end-to-end data control. EdgeClaw handles simple tasks locally with zero token consumption, while complex tasks are intelligently routed to the cloud. It is already compatible with hardware such as Pinecone Pi and NVIDIA DGX Spark, and the EdgeClaw Box is available out-of-the-box hardware, allowing non-technical users to easily have secure and controllable AI digital employees.

EdgeClaw's main functions

• Intelligent agent execution frameworkBased on the OpenClaw architecture optimization, it supports the secure operation of AI agents on local devices without relying on the cloud.
• Three-layer data security protocolThrough tiered protection of S1 (public information uploaded to the cloud), S2 (sensitive information anonymized), and S3 (privacy data locked locally), data security and controllability are ensured.
• Edge-cloud collaborative computingSimple tasks are processed locally (zero token consumption), while complex tasks are automatically routed to the cloud, balancing performance and cost.
• Dual-track memory mechanismMEMORY.md (an anonymized version for cloud use) and MEMORY-FULL.md (a full version for local use only) enable hierarchical data management.
• Local model supportThe integrated wall-mounted MiniCPM edge model can still run in offline environments, ensuring privacy and availability.
• GuardAgent ProtocolA six-stage lifecycle hook system enables full-process control of sensitive data.

Key information and usage requirements of EdgeClaw

• DeveloperWallfacer Intelligence, Tsinghua University THUNLP, Renmin University of China, AI9Stars, OpenBMB
• licenseMIT Open Source License
• technology stack: TypeScript (83.3%), Swift (12.7%), Kotlin (1.8%)
• Core positioningSupports edge-cloud collaborative AI frameworks where both models and agents can be deployed locally.
• Hardware RequirementsSupports mainstream edge devices such as Pinecone Pi, NVIDIA DGX Spark, and Apple Mac Mini, with optional EdgeClaw Box hardware available out of the box.
• Software environmentRequires configuring a local runtime environment (simplifies the deployment process compared to OpenClaw), and supports integration with edge models such as MiniCPM for offline operation.

EdgeClaw's core advantages

• Safe and controllableIt pioneered a three-layer data security protocol (S1/S2/S3) to enable public information to be uploaded to the cloud, sensitive data to be de-identified, and private content to be locked locally. Combined with GuardAgent's six-stage lifecycle management, it addresses the risk of data leakage at the source.
• End-to-cloud collaborationSimple tasks consume zero tokens locally, while complex tasks are intelligently routed to the cloud, balancing privacy protection and performance requirements, with marginal costs approaching zero.
• Local priorityIt integrates the MiniCPM edge model, ensuring availability even when the network is down, eliminating concerns about network fluctuations or service interruptions, and truly enabling autonomous control over data sovereignty.
• Ready to use right out of the boxThe framework simplifies the deployment process and is compatible with mainstream edge hardware, leading to the launch of the EdgeClaw Box hardware product.
• Cost advantageIt supports local execution of daily tasks, reducing costs by several orders of magnitude compared to pure cloud solutions, making it particularly suitable for high-frequency, high-privacy scenarios.

How to use EdgeClaw

• Environmental preparationInstall the EdgeClaw framework and configure the local runtime environment. Choose an adaptation solution based on your hardware: edge devices such as Pinecone Pi, NVIDIA DGX Spark, and Mac Mini, or purchase the EdgeClaw Box hardware out of the box for immediate use.
• Model DeploymentDownload and integrate the MiniCPM client-side model to ensure local offline operation capability. Configure the cloud model interface as needed for handling complex tasks.
• Security ConfigurationEnable a three-tier data security protocol: set S1/S2/S3 data classification rules, configure GuardAgent lifecycle hooks, and establish a dual-track memory mechanism (MEMORY.md anonymized version and MEMORY-FULL.md full version).
• Start runningStart the local Gateway to maintain intranet access (avoid direct exposure to the public network). Call the intelligent agent via API or interface; the system automatically determines the task complexity and handles it locally or in collaboration with the cloud.

EdgeClaw project address

• GitHub repositoryhttps://github.com/OpenBMB/EdgeClaw

EdgeClaw's Competitive Product Comparison

product	Developer	Core differences	Advantages	Disadvantages
EdgeClaw	Wall-facing Intelligence / Tsinghua University	End-to-cloud collaboration, three-layer security protocol	Data tiered protection, local priority, low cost, and availability even when offline.	Ecological maturity needs to be improved
OpenClaw	Anthropic	Cloud-first remote control framework	Feature-rich and ecologically sound	Data needs to be uploaded to the cloud, which poses a risk of leakage.
Tencent QClaw	Tencent	WeChat ecosystem integration, minimally simplified packaging	WeChat mini programs offer easy access, low barriers to entry, and are suitable for domestic users.	Relying on the WeChat ecosystem, data privacy depends on Tencent.
JVS Claw	Alibaba	Cloud-based AI Agent Platform	Ample free credits and strong cloud computing power.	Pure cloud solution, data leaves local storage

Application Scenarios of EdgeClaw

• Personal privacy protectionIt supports local processing of private data such as chat logs, photos, and documents, ensuring that sensitive information is not sent to the cloud and preventing the risk of cloud leakage.
• Enterprise data securityIndustries such as finance, healthcare, and law handle confidential information such as customer data, medical records, and contracts to meet compliance requirements and achieve independent control over data sovereignty.
• One-person company (OPC)Entrepreneurs can use EdgeClaw Box to build their own digital employee teams and complete daily tasks such as document processing, schedule management, and content creation at low cost.
• Offline environment operationIn situations such as field research, remote areas, and unstable network environments, the device can independently complete tasks such as data analysis and report generation using edge-side models.