How ARCA SENTRY works
All the conceptual + technical material in one place. Use this for demos and walkthroughs.
Connect any agent
Onboarding takes ~30 seconds. SENTRY supports drop-in proxy mode (OpenAI / Anthropic / Gemini SDKs), raw HTTP endpoint testing, web chat-widget scanning, and (soon) WhatsApp + Facebook Messenger.
Proxy mode · 1 line of code
Point your OpenAI / Anthropic / Gemini client at SENTRY's URL instead of the upstream provider. Every call is audited in flight. Critical violations are blocked at the gateway with HTTP 451; warnings pass through with diagnostic headers. See the Proxy & SDKs tab for code snippets.
Auto-Fix · don't block the user, rewrite the response
When the audit detects a violation, SENTRY can ask Gemini Pro to rewrite the response so the end user receives a compliant reply instead of an error. The original AND the rewrite are both stored — your compliance team reviews the diff, not a customer complaint.
Red Team · automated pen-testing
Run the attack suite against any registered agent. 12+ pre-built attacks across the 5 regulation families — credit-decision opacity, GDPR right-to-explanation, DORA incident hiding, PII probes, prompt-injection chains. Output: pass/fail score, breakdown, downloadable PDF report.