Open Source, Confidential by Design: The Linux Stack for Agentic AI
summary
Claims a Linux-first agent OS (Eliza OS), an agent harness wired to Gmail/Signal/Telegram that accidentally replied to a friend, and a demo agent inside a TDX machine that verifiably deleted its credentials for seven days. Shows practical tooling: DStack on GitHub and mentions running Quinn 3.6 31B locally on a 48 GB MacBook. Describes structured/constrained generation, TE escape via radio-wave side channels, and role-based Linux permissions for agent owner/agent/admin separation.
AI-written summary · about
chapters (8)
- 0:00Confidential AI infrastructure
- 1:06TEE use cases
- 7:09Agentic OS on Linux
- 10:40LLM sampling research
- 15:45Multi-agent coordination
- 23:18Agent security challenges
- 25:46Deterministic agent workflows
- 29:39TEEs and agentic AI
related talks
34mOpen Source AI Agents on User-Owned Infra: K3s, MCP, & GPU Sharing in Practice
19mKeynote: Towards Building Safe and Secure Agentic AI
40mAgents in Action: Advancing Open Source AI Missions
46mAI Agents Are Becoming Operating Systems: Lessons From OpenClaw
15mBuilding Open AGI Loyal to Humanity
17mHow to Make Useful AI Agents
37mAGENTS OF S.E.A.L.E.D: AI Agentic Cybersecurity Framework
24mLightning Talk: Building the AIverse: Standardizing Autonomous AI Agents for a Decent...