Presentation: Enchant Your AI and APIs with eBPF Magic πͺ
Dan Finneranμ eBPFλ₯Ό μ¬μ©νμ¬ AI API νΈλν½μ μ μ΄νλ λ°©λ²μ μ€λͺ ν©λλ€.
Dan Finneranμ μμ° νκ²½μμ κ΄λ¦¬λμ§ μλ AI μμ± μ½λμ μνμ λ Όμν©λλ€. κ·Έλ eBPFκ° Kubernetesμμ AI API νΈλν½μ κ°λ‘μ±κ³ μ μ΄νλ λ°©λ²μ 보μ¬μ€λλ€. 컀λ μμ€μ μμΌ ν μ ν΅ν΄ ν¬λͺ ν ν둬ννΈ νν°λ§, λͺ¨λΈ κ΅μ²΄, ν ν° μ ν, μμ€ν νΈμΆ μ νμ κ°λ₯νκ² νμ¬ μ ν리μΌμ΄μ μμ€ μ½λλ₯Ό μμ νκ±°λ 컨ν μ΄λλ₯Ό μ¬μμνμ§ μκ³ AI μμ΄μ νΈλ₯Ό 보νΈν μ μμμ μ€λͺ ν©λλ€.
Dan Finneran discusses using eBPF to manage AI API traffic in production.
Dan Finneran addresses the risks of unmonitored AI-generated code in production environments. He demonstrates how eBPF can intercept and control AI API traffic in Kubernetes. By utilizing kernel-level socket hooks, he explains how prompt filtering, model swapping, token limits, and syscall restrictions can secure AI agents without needing to modify application source code or restart containers.