Show HN: Talos – An AI agent with a permission kernel between model and shell
Summary
Talos presents an AI agent framework with a permission kernel that gates every tool call to a browser-based shell. It emphasizes deterministic, per-action authorization, sandboxed execution, and multi-channel communication, aiming to prevent model misuse while enabling autonomous tasks. The project provides a detailed manifest of 23 tools, an adversarial/testing narrative, and strict provenance controls, all under MIT license.