Spanlens

Spanlens provides comprehensive visibility for your language model interactions. With a single line of code, you can effortlessly track: • Request logging with full details (cost, latency, tokens) • Multi-step agent tracing with critical path analysis • Anomaly detection for latency and cost • PII scanning and data protection • Model-swap suggestions for cost optimization This robust platform captures every call from providers like OpenAI, Anthropic, Gemini, Azure, and Ollama, offering deep insights into performance and expenditure. You gain granular per-request breakdowns, daily cost rollups, and proactive budget alerts, allowing you to understand and control your spending before the bill arrives. The agent tracing capability visualizes complex, multi-step workflows as intuitive waterfall span trees, making it easy to pinpoint performance bottlenecks and optimize execution paths. Spanlens also incorporates advanced features such as real-time anomaly alerts for significant deviations in latency or cost, PII scanning on request bodies with automatic API key masking, and insightful model recommendations to help you achieve better performance at a lower cost. Furthermore, it supports evaluating response quality with provider-as-judge scores and replaying datasets across different prompt versions and models for comparative analysis. Built for developers and engineering teams, Spanlens is MIT-licensed, self-hostable, and offers a free tier designed to support growth. It's an essential tool for anyone needing to monitor, optimize, and secure their language model applications without added complexity.