# JoySafety/JoySafety

JoySafety

Repository: https://github.com/JoySafety/JoySafety
Canonical: https://ross.abutalabs.com/products/joysafety
Language: Java
License: NOASSERTION
License Family: other
Last push: 2026-05-11T13:04:08+00:00

## Health v2 (maintenance only)
Score: 47/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 81, release rhythm 17, longevity 24
- inputs: {"age_days": 343, "days_push": 114, "days_rel": 342, "gap_med": null, "n_releases_24m": 1}
- flags: no_license
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1344, forks 94 (observed 2026-08-28T04:04:27.207031+00:00)

## What it is
JoySafety is an open-source large language model safety framework from JD.com, written in Java, providing prompt injection detection, content moderation, and multi-turn conversation risk identification. It uses a pluggable DAG-based orchestration of detection models (BERT/FastText/Transformer) with hot-updatable policies and supports Docker Compose deployment at production scale.

## Use cases
- detect prompt injection attacks against LLM applications
- moderate chatbot responses for policy compliance
- guard enterprise AI assistants against data leakage
- block unsafe LLM outputs in production
- orchestrate multiple safety detection models with DAG policies
- add content safety guardrails to RAG pipelines
- self-host an LLM content safety service

## When to choose
- you need production-grade, battle-tested LLM safety guardrails with high throughput
- you want hot-swappable detection models and minute-level policy updates
- you deploy Java/JVM services and want Docker Compose-based self-hosting
- you need multi-turn conversation risk detection and red-line knowledge base responses

## When to avoid
- you need a lightweight client-side or embedded guardrail library rather than a deployed service
- your stack cannot run JVM services or Docker
- you need English-only moderation with no Chinese-language focus
- you require a permissive standard open-source license (license is non-standard)

## Facets
- artifact type: framework
- maturity: active
- function: security, nlp, machine-learning, middleware, api-framework, rag
- domain: large-language-models, security, artificial-intelligence, chatbots, self-hosted
- platform: self-hosted, jvm
- tags: llm-safety, prompt-injection-detection, content-moderation, guardrails, dag-orchestration, jd, chinese-nlp, hot-reload, docker-compose, natural-language-processing, docker, linux, web-server

## Member repositories
- JoySafety/JoySafety (main) score 47

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:04:27.207031+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T04:42:48.209275+00:00, confidence not recorded.
  - readme: https://github.com/JoySafety/JoySafety (fetched 2026-08-28T04:04:27.207031+00:00, sha ce956feb574f)
- Data as of 2026-08-30T08:39:29.467469+00:00.