F5 AI Guardrails out-of-band

AI inference networking  
Part 1 Open WebUI as front door for ChatGPT
Part 2 Inference security using Open WebUI
Part 3 Layering NGINX for more Inference security
Part 4 F5 AI Guardrails out-of-band
Moving to out-of-band model for prompt security

Why put AI guardrails out-of-band?

I’m aiming for a model-agnostic approach in AI. Actually I’d like to be agnostic on almost everything. By running F5 AI Guardrails out-of-band, I plan to move some orchestration decisions back to my Open WebUI instance.

For example, I want to be able to do things like this:

  • Multiple models
    • I may switch models easily from Open WebUI without touching F5 AI Guardrails
    • Perhaps if Guardrails flags a prompt or response, I’ll have options at Open WebUI. Examples:
      • I could send a “blocked” prompt to an on-prem model
      • I could try my hand at redacting offending text from a flagged response
  • Policy enforcement centralized
    • depending on which scanner blocks a prompt or a response, I may or may not honor the decision of Guardrails
    • I could bypass the Guardrails checks selectively
  • etc

Pre-requisites

  • Follow blog posts 1, 2 & 3. By now you should be able to deploy and destroy our environment on demand.

Function for out-of-band prompt/response inspection

Conclusion

Updated: