Metadata-Version: 2.4
Name: haetae
Version: 0.0.1
Summary: Robot safety & security stack for physical AI: a policy gate between untrusted models and actuators (pre-alpha)
License: Apache-2.0
Project-URL: Repository, https://github.com/haetae-robotics/haetae
Keywords: robotics,safety,security,ros2,guardrail
Classifier: Development Status :: 1 - Planning
Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
Classifier: Topic :: Security
Classifier: License :: OSI Approved :: Apache Software License
Requires-Python: >=3.9
Description-Content-Type: text/markdown
License-File: LICENSE
Dynamic: license-file

# Haetae (해태) /hɛ.tʰɛ/ "heh-teh"

**Robot safety & security stack for physical AI.**

Haetae treats VLA / foundation-model output as *untrusted input*. Every proposed
robot action passes through a single policy gate that returns one of three verdicts:

| Verdict | Hanja | Meaning |
|---|---|---|
| `yun`  | 允 | allow |
| `jeol` | 節 | clamp (allow with reduced speed/force/workspace) |
| `bul`  | 不 | deny |

Planned components: `sillok` (tamper-evident incident recorder), `maek` (self-diagnosis),
`jangseung` (physical-space consent), `amhaeng` (red-team harness).

> Status: **pre-alpha (0.0.x)**. Not a certified safety device.
