Galileo

Galileo provides an enterprise-focused GenAI Studio for evaluating, observing, and protecting generative AI systems in development and production. It centralizes evaluation metrics, real-time observability, and runtime guardrails so teams can iterate on prompts, models, and retrieval strategies with measurable feedback. Galileo runs low-latency evaluators (Luna models), hosts inference for live monitoring, and offers SDKs and APIs to integrate logging and traces into existing applications.

• Evaluate — Rapidly run and compare combinations of prompts, models, embedding params, and chain nodes to find the right configuration; use the platform UI or SDKs to log experiments and golden test sets.
• Observe — Monitor live traffic, traces, and sessions with pre-built and custom metrics to detect drift, latency, and accuracy issues in production systems; connect via SDKs or API to stream logs and traces.
• Protect — Intercept requests and responses in real time with guardrail policies and threat detection to block harmful outputs or attacks before they reach users.
• Luna evaluation models — Use Galileo’s low-latency evaluator models to run automated judgements (<200ms typical) for production monitoring and inexpensive continuous evaluation.
• SDKs & integrations — Install Python or TypeScript SDKs, initialize with an API key, and add the log decorator or GalileoLogger to capture prompts, responses, traces, and spans.

Getting started: sign up for the hosted console, install the Python or TypeScript SDK, set GALILEO_API_KEY, and either use the log decorator or GalileoLogger to begin sending traces and running evaluations.