Galileo

Galileo is a web and API observability and evaluation platform for enterprise AI teams building, deploying, and operating generative AI applications and agents.

It supports the full eval-to-guardrail lifecycle so teams can debug issues in development and enforce policies in production.

• Agent observability with framework-agnostic tracing via a Graph Engine, plus Timeline and Conversation views to inspect multi-step, multi-agent behavior.
• Evaluations and experiments to score LLM outputs using out-of-the-box evaluators for RAG, agents, safety, and security, along with custom metrics that reflect domain requirements.
• Production monitoring for prompts and responses, token usage, latency, behavior, and health signals, with alerts for anomalies.
• Galileo Protect guardrails to help block prompt injection, PII leakage, hallucinations, and harmful content at runtime.

Pricing: Free Agent Reliability Platform is available; additional pricing details are not provided.

Notable for turning offline evaluation work into production controls, including distilling LLM-as-judge evaluators into Luna-2 small language models for lower-latency, lower-cost evaluation at scale, plus native support for frameworks such as LangChain, CrewAI, and the OpenAI SDK (with integrations including Zuplo AI Gateway).