# Glass > Glass is an AI agent observability and improvement platform by XenosLabs. It helps teams observe, classify anomalies, build evals, and optimize prompts for their AI agents — automatically. Glass detects silent failures that traditional monitoring tools miss: hallucinations, inaccurate responses, redundant tool calls, context window overflow, broken tool outputs, infinite loops, missed user intents, stale data retrieval, prompt injection leaks, guardrail bypasses, latency spikes, and token waste. The platform provides a continuous improvement flywheel: trace every LLM call and decision, surface anomalies with AI-powered classification, build evaluations from real failures, and optimize prompts automatically. Glass integrates via a lightweight SDK (Python, TypeScript) with a few lines of code and supports OpenAI, Anthropic, and other major LLM providers. ## Docs - [Homepage](https://glasshq.ai/): Product overview, features, and getting started - [Pricing](https://glasshq.ai/pricing): Free tier (~20k traces/mo), Standard ($299/mo), and Premium plans - [About](https://glasshq.ai/about): Team, mission, and company background - [Blog](https://glasshq.ai/blog): Articles on AI agent monitoring and best practices ## App - [Sign Up](https://app.glasshq.ai/signup): Create a free Glass account - [Dashboard](https://app.glasshq.ai/dashboard): Real-time metrics, latency, success rates, token usage, and costs - [Login](https://app.glasshq.ai/login): Access your Glass dashboard ## Key Features - [Trace Monitoring](https://glasshq.ai/#features): Track every LLM call, tool execution, decision, and cost in a span tree view - [Anomaly Detection](https://glasshq.ai/#features): AI-powered classification of silent failures across 12+ failure categories - [Evaluation Builder](https://glasshq.ai/#features): Build evals from real production failures to prevent regressions - [Prompt Optimization](https://glasshq.ai/#features): Automatically improve prompts based on observed failure patterns - [Alerting](https://glasshq.ai/#features): Get notified via Slack, email, Google Chat, or Microsoft Teams when issues arise