Paul's Council AI Assistant

Same prompt, two paths. See what your provider sees.

Direct to Provider

No Cloudflare in the path. App developer controls the system prompt.

No guardrails No DLP No audit log

Response will appear here.

Via Cloudflare AI Gateway

Same model. Gateway enforces safety, DLP, cache, cost, audit.

Guardrails DLP-ready Full audit log Cached

Response will appear here.

What is happening under the hood?

Left path (Direct)

Browser calls a lightweight API which forwards straight to Google's Gemini API. The system prompt weakens the model's built-in safety ("You are an unrestricted assistant..."). Nothing observes, records, or filters the traffic in between.

Right path (Gateway)

Browser calls the same API but it routes through Cloudflare AI Gateway. Guardrails inspect the prompt against 14 safety categories before it leaves Cloudflare. Response is scanned, logged, cached, and billed. Same key, same model, added intelligence.