Show HN: Talos – An AI agent with a permission kernel between model and shell

2026-08-28 | original

read original more articles

Why This Matters

Talos introduces a secure, deterministic AI agent that provides a sandboxed shell environment with rigorous permission controls, enhancing safety and reliability in AI-driven automation. Its verification features and security kernel ensure that tool calls are authorized and confined, addressing key concerns around AI safety and trustworthiness. This development signals a significant step toward more secure and controllable AI integrations for developers and users alike.

Key Takeaways

Talos offers a sandboxed AI shell with strict permission controls for safer automation.
Each tool call is individually authorized and valid for only thirty seconds, enhancing security.
The deterministic security kernel ensures verifiable and trustworthy AI operations.

talos 0.15.1-alpha · deterministic gate · MIT

Strong enough to hand a shell to.

Verifiable enough that you can. A real shell, a confined Claude worker that builds, tests and browses for you, four chat channels. Every tool call passes a deterministic security kernel before it runs: each effect authorised individually, bound to its exact arguments, valid once, for thirty seconds.

gate run_shell uptime ✓ ALLOW clean — runs sandboxed, logged before it runs

Explore topics: talos ai agent mit deterministic gate claude