HiveTraceLab

Welcome to HiveTrace!

LLM monitoring, guardrails, and red teaming for production GenAI applications.

Website (EN) · Website (RU) · HiveTrace Red · Docs · GitHub · Telegram

Products

• HiveTrace — real-time LLM monitoring platform. Bidirectional analysis of user ↔ LLM traffic: prompt injections, jailbreaks, PII leaks, anomalous behavior.
• HiveTrace Red — automated red team platform for adversarial testing of GenAI systems.
• Red Team Audit — expert-led security assessment service.

Open Source

Guardrail models

• HiveTraceGuard-Pro (0.6B) — compact guardrail model for detecting harmful content, jailbreaks, prompt injections, and agentic threats.

📄 Technical Report: arXiv:2609.01046

GLiNER Guard · collection

• gliner-guard-omni (0.3B) — single forward pass: safety, PII (32 types), adversarial attacks (15 labels), harmful content (30 labels), intent, tone. Zero-shot custom policies via natural language.
• gliner-guard-uniencoder (0.1B) — edge deployment.
• gliner-guard-biencoder (0.1B) — edge deployment, bi-encoder architecture.

📄 Technical Report: arXiv:2605.05277 · Live demo

Benchmarks & Datasets

• GuardRate Leaderboard — interactive leaderboard comparing guardrail models by safety performance, latency, and multilingual capabilities.
• strongrejectPlusPlus — StrongREJECT translated into 4 languages including Russian + original English version.
• pii-bench — PII detection benchmark for evaluating guardrail models.