Galileo
AI evaluation and observability platform, now part of Cisco and Splunk, that turns offline evals into production guardrails with its Luna models
Overview
Galileo ("Don't just monitor AI failures. Stop them.") is an AI observability and eval engineering platform: you build evaluations for RAG apps and agents offline, then run the same checks in production as guardrails that can block a response, limit an agent's tool access or escalate to a human. It ships 20+ ready-made evaluators for RAG, agents, safety and security, lets you write custom ones, and runs them cheaply with Luna, its small distilled evaluation models. Insights analyzes failure modes and suggests fixes, and it deploys as SaaS, in your VPC or on-prem. Cisco completed its acquisition of Galileo on May 22, 2026 and is folding it into Splunk Observability Cloud, where it is now sold as Splunk Agent Observability. Galileo also maintains an open-source Agent Leaderboard that ranks LLMs on agentic tasks. Before the deal it raised a $45M Series B (October 2024) led by Scale Venture Partners. Not to be confused with Galileo AI, the UI design startup Google folded into Stitch. Alternatives include Braintrust, Patronus AI and Arize.
The Verdict
Who Should Use Galileo?
Best For
- Enterprises running agents and RAG apps in production
- Teams that want the same evals offline and as live guardrails
- Splunk and Cisco customers adding AI observability
Not Ideal For
- Teams that want a fully open-source eval stack
- Buyers wary of a product mid-way through an acquisition integration
What's Great
- Evals become production guardrails without rewriting them
- Luna models make it cheap to score all traffic (vendor claim)
- SaaS, VPC or on-prem deployment
Watch Out For
- Pricing and plans now live on Splunk, billed annually
- Roadmap now tied to Splunk Observability Cloud
- Few independent reviews; most comparisons are written by rivals
Pricing
View all features & details
Key Features
- 20+ evaluators for RAG, agents, safety and security
- Custom evaluators and LLM-as-judge
- Luna distilled evaluation models
- Guardrails that block, restrict tools or escalate
- Insights for failure analysis
Company
- Co-founders include Vikram Chatterji (CEO) and Atindriyo Sanyal
- Series B $45M (Oct 2024), Scale Venture Partners
- Acquired by Cisco, closed May 22, 2026
- Open-source Agent Leaderboard on GitHub and Hugging Face
How It Compares
| Feature | Galileo | Braintrust | Patronus AI |
|---|---|---|---|
| Main strength | Evals that run as live guardrails | Eval-first observability | Automated testing and hallucination detection |
| Own small eval models | Yes (Luna) | See vendor | See vendor |
| On-prem | Yes | See vendor | See vendor |
| Owner | Cisco (Splunk) | Independent | Independent |
Sources: Galileo · Cisco acquisition page · Series B (Oct 2024) · Agent Leaderboard
Is this your tool? Add a badge