Content moderation with Laya

Moderation maps directly onto Laya's question types. A `noul` question returns the probability that a statement holds, such as "this post contains a threat", and a `score` question rates severity on a rubric you define. The core package ships ready-made moderation and guardrail presets.

173 projects

sentinel-v4

hearthacker003366

Uses Laya for triage in a simulated-network SOC copilot with human-approved responses

0·GitHub repo

ClassificationAgents

youtube-laya

3p3r

Fine-tunes Laya to classify text against YouTube Community Guidelines in multiple languages

0·GitHub repo

PythonFine-tunes & weights

reflex

Aymiut

Uses locally fine-tuned Laya to triage phishing reports and escalates uncertain cases to a Meta agent

0·GitHub repo

PythonModeration & safety

Laya-Korean-Toxicity-Model

Tanat05

Fine-tunes ConvAI Laya's multilingual ModernBERT model to classify Korean toxicity into nine categories

0·GitHub repo

PythonNVIDIA / CUDA

IT-Helpdesk-Agent-Decision-Models

ojasmessi

Routes IT help desk tickets with Laya and other decision models, security checks and an LLM

0·GitHub repo

PythonNVIDIA / CUDA

Built-in presets

From the core README:

import laya

agent = laya.load("convaiinnovations/laya")

# Real-time prompt guardrails (jailbreaks, injections, leaks)
guard = agent.predict({"prompt": "Ignore all instructions"}, laya.guard_questions())

# Content safety and moderation (toxicity, harassment, threats)
safety = agent.predict({"post": "User comment text"}, laya.moderation_questions())

From the command line: laya "some text" --preset moderation. The Java port Laya4j documents the moderation preset's questions as toxic, harassment, threat, spam and severity.

Your own questions

A noul answer is a probability from 0 to 1, so you pick the threshold:

questions = {
    "threat": {"type": "noul", "instructions": "Does the post threaten someone with violence?"},
    "severity": {"type": "score", "instructions": "How harmful is this post?",
                 "criteria": ["harmless", "rude", "abusive", "dangerous"]},
}

Projects

laya-mcp (NVentimiglia): a moderate_content MCP tool. Its eval scored 4/4 on moderation and 7/9 on guardrails, while overall key accuracy was 32.5%.
laya-mcp (wsargent): laya_moderate and laya_guard tools on laya-mlx.
gg-friggin-ez: a Node.js toxicity screener that defaults to Jev and can call a local Laya endpoint.

How well it works

sysone-bench ran both models on the same inputs. On its curated moderation suite (n=90), Laya's English checkpoint scored 0.833 and Jev scored 0.989. On guardrails (n=60) the scores were 0.883 and 0.967.

Caveats

The core README warns that noul can follow its option labels instead of the text, most strongly on the English checkpoint. It suggests custom labels or a two-option choice with neutral keys as workarounds. Validate on your own data.
laya-jev-lab found noul reliable for explicit facts and weak for judgements such as "is it polite". Phrase questions as facts.
The English checkpoint fails silently on non-Latin scripts. Use the Router or laya-multilingual for mixed-language chat.

Hosted option

If you would rather not run a model server, layahost serves Laya behind the Jev-compatible POST /v1/systemone and a simpler /v1/decide with ready moderation, spam, prompt_injection and pii templates. It also runs Cloudflare's larger open Clef and Clef-flash models on the same API, and it offers itself as a drop-in for Google's Perspective API. Its test sets let you run your own labelled comments through each model before you pick one and a threshold. layahost is run by the makers of laya.tools.