junyul-kr 1.2.2

Junyul — Korea AI Framework Act compliance SDK for Python. 3-line integration for automatic disclosure, watermarking, compute tracking, and audit-ready events.

junyul-kr

Korea AI Framework Act (AI 기본법) compliance SDK for Python.

Install

# Public PyPI release or approved enterprise package channel
pip install junyul-kr
# With extras:
pip install "junyul-kr[image,audio,pytorch]"

Until the public PyPI release is opened, enterprise customers install from the approved private package channel or the signed wheel/sdist supplied through the Junyul adoption channel.

Three-line integration

import junyul

junyul.init(api_key="JUN_live_xxx", environment="production")

@junyul.track(asset_id="gpt_chatbot_v1")
def ask_chatbot(user_question: str) -> str:
    return openai_client.chat.completions.create(...)

Features

• @junyul.track(...) — automatic event recording (input/output hash, latency, tokens)
• Failed tracked calls are recorded with hashed error metadata and no raw error message.
• junyul.events.record_objection / record_human_review
• junyul.events.record_heartbeat(asset_id=...) — SDK queue/outbox/circuit telemetry
• junyul.classify(...) — live classification API
• junyul.watermark.image / video / audio / text
• junyul.compute.ComputeTracker — PyTorch/JAX/TF hooks for 10^26 FLOPs tracking
• FastAPI / Flask / Django / Starlette middleware for auto-disclosure injection
• Test mode: junyul.init(environment="test"); junyul.testing.get_captured_events()

See the plan doc for the full design.