> ## Documentation Index > Fetch the complete documentation index at: https://docs.visceralai.dev/llms.txt > Use this file to discover all available pages before exploring further. # Visceral > A performance engine for AI agents — it makes wasted LLM calls cheaper, but only once it can prove the cheaper call is safe. Visceral wraps your agent, watches its LLM traffic, finds where work is wasted, and makes that call cheaper — **but only once it can prove the cheaper call is safe.** It is a performance engine, not a high-level router: it surfaces an optimization only when it can prove that optimization is safe, and it never makes anything worse. Two rules sit above everything we do: An optimization ships only when Visceral can prove it is output-preserving. If it can't prove it, it doesn't touch the call. If Visceral is down or degraded, your agent runs exactly as it would without us. We can never be the reason your agent breaks. ## Get started Instrument your agent and see its traffic in a few lines. The gate, the DeltaGraph, and why we only optimize what we can prove. Add Visceral to a Python agent. What we store (metadata only) and how it's isolated and encrypted.