---
title: HiveTraceLab
colorFrom: blue
colorTo: pink
sdk: static
pinned: true
---

# HiveTraceLab

Welcome to HiveTrace!

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

[Website (EN)](https://hivetrace.ai/?from=hf) · [Website (RU)](https://hivetrace.ru/?from=hf) · [HiveTrace Red](https://hivetrace.ru/red/?from=hf) · [Docs](https://hivetrace.github.io/docs/en/?from=hf) · [GitHub](https://github.com/HiveTrace/?from=hf) · [Telegram](https://t.me/aisecuritylab/?from=hf)

## 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
**GLiNER Guard** · [collection](https://huggingface.co/collections/hivetrace/gliner-guard-v1)

- `gliner-guard-omni` (0.3B) — flagship. 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.

[arXiv:2605.05277](https://arxiv.org/abs/2605.05277) · [Live demo](https://huggingface.co/spaces/hivetrace/gliner-guard-demo)

### Datasets

- [`strongrejectPlusPlus`](https://huggingface.co/datasets/hivetrace/strongrejectPlusPlus) — StrongREJECT translated into 4 languages including Russian + original English version.
- [`pii-bench`](https://huggingface.co/datasets/hivetrace/pii-bench) — PII detection benchmark for evaluating guardrail models.