Qingteng WorkClaw

WorkClaw is an enterprise management platform developed by Qingteng for governing OpenClaw-based AI agents within organizations. It centralizes control over agent operations to prevent sensitive data from leaving the organization and to manage risks associated with autonomous AI agent execution. The product addresses several risk areas identified by the vendor: - API keys stored in plaintext on endpoints, with no sensitive data masking - Weak prompt injection defenses, where a large share of attacks can bypass protection - Unreviewed third-party components/Skills that can introduce malicious code - Lack of access controls, allowing agents unrestricted calls to shell, files, and network resources - Decentralized, ungoverned agent deployment by employees with no unified oversight - Untracked token consumption and unmanaged employee-configured assets - Lack of audit/traceability for agent actions - No standardized evaluation process to detect quality regressions when models or prompts change Core functional areas: - Security Control: full lifecycle AI security governance across data, models, tools, and execution; unified entry-point control so raw sensitive data does not reach the LLM; centralized API key management with sensitivity-based routing to internal/external models; quota and rate limiting; code scanning and security review of third-party Skills; fine-grained tool access permissions including command interception - Unified Management: lifecycle management from employee onboarding to offboarding; visual dashboards for instance monitoring; token quota allocation and usage statistics with auto rate-limiting; policy templates by department/role; rapid (30-second) agent instance deployment; behavior auditing with real-time alerts on high-risk actions; a "Skill store" for distributing vetted tools; cross-device sync of configurations and history - Continuous Evaluation: automated evaluation triggered by prompt/Skill/model changes; evaluation datasets