TruClaw - Biometric Guardrails for Google ADK Agents

Autonomous agents, especially long-running ones with large context windows, can perform irrevocable harmful actions. The causes are varied:
1. Misinterpreting security policies & guardrails
2. Hallucinating harmful actions
3. Prompt injection
4. Agent drift influenced by large context windoes
5. Agent takeover
Monitoring & observing do not sufice - we need a hard stop. TruClaw is this agentic killswitch. We have developed custom ML models running 100% on edge mobile devices. These models verify its your face authorizing agent actions. No biometric leaves the device. Only hardware attested authorization proof is delivered in responses.
TruClaw uses Gemini to observer and understand the semantic intent of ever agent action. Ones that are deemed suspicious trigger a mobile push notification for TruClaw OnDevice authorization.
This is a bold,new step in Agentic Security. Its open source with generous licensing for free trials.

Видео TruClaw - Biometric Guardrails for Google ADK Agents канала Sanjay Krishnamurthy