Introduction

AI Runtime governance platform that monitors, detects, and provides evidence of AI risk in production

Avaliar is an AI runtime governance platform that monitors, detects, and provides evidence of AI risk in production LLM applications. Whether you are dealing with hallucinations, bias, toxicity, PII leaks, or prompt injection attempts, Avaliar provides real-time detection and actionable insights to keep your AI systems safe and reliable.

Quickstart

Get up and running with Avaliar in 5 minutes.

Python SDK

Install and configure the Avaliar Python SDK.

Proxy Integration

Route LLM traffic through the Avaliar Proxy for zero-code tracing.

How It Works

Avaliar follows a three-stage lifecycle that takes every LLM interaction from raw data to resolved insight.

1. Capture

   Instrument your application with the @traceable decorator or route traffic through the Avaliar Proxy. Every request and response is recorded as a structured trace with full metadata, including model, provider, latency, and token counts.

2. Detect

   Traces are automatically passed through the detection pipeline. Ten built-in detectors analyze every input and output for safety issues — prompt injection (covering jailbreaks and SQL injection), toxicity, PII, hallucination, bias, misinformation, misuse, graphic content, illegal activities, and personal safety.

3. Alert

   When a detector finds an issue, Avaliar scores its severity and evaluates it against your alert rules. Notifications are dispatched to your configured channels so your team can investigate and respond immediately.

Explore the Platform

Core Concepts

Understand traces, spans, detectors, issues, alerts, and reports — the building blocks of Avaliar.

Platform

Explore the Traces dashboard, analytics, alerts, and reports in the Avaliar platform.