Veil AI Firewall protects prompts, responses, and MCP tool calls with built-in PII redaction, prompt injection detection, output filtering, and hallucination flags. Keep your SDK. Change one URL.
Summarize this support ticket: "Hi, my name is Sarah Johnson and my email is sarah.j@company.com. My SSN is 078-05-1120 and I'm calling from 555-867-5309. My card 4111-1111-1111-1111 was charged twice."
Summarize this support ticket: "Hi, my name is <<VEIL_PERSON_a8f2c3d1e4f5>> and my email is <<VEIL_EMAIL_ADDRESS_c3d1e4f5a8b2>>. My SSN is <<VEIL_US_SSN_9e7b1a2c3d4e>> and I'm calling from <<VEIL_PHONE_NUMBER_4f2a1b3c8d9e>>. My card <<VEIL_CREDIT_CARD_b1e8a2c3d4f5>> was charged twice."
Try the built-in redaction layer on a real prompt. Then turn on runtime input, output, and MCP controls in the API.
Keep your current SDK, models, and provider keys. Point your client at Veil and turn protections on with headers or standalone firewall endpoints.
Veil still handles privacy. Now it also gives you runtime controls for the two other places AI apps break: model behavior and tool surfaces.
Redact PII, catch prompt injection, and stop obvious approval-bypass or credential-harvesting attempts before they reach the model.
Flag risky responses, prompt leakage, suspicious links, unsafe tool arguments, and unsupported new claims before they leave your stack.
Inspect tool descriptors, calls, and results for hidden instructions, scope mismatches, destructive actions, and data exfiltration paths.
Same auth flow, same billing, same provider routing. You can add runtime security without migrating off the Veil integration you already have.
Keep the default proxy behavior backward compatible, then opt into input blocking, output monitoring, or hallucination flags per request.
Veil is being built from real vulnerability research and public MCP audits, not generic “guardrails” copy disconnected from how attacks actually look.
41 named providers. Veil AI Firewall protects the traffic without forcing a model-vendor rewrite.
PII redaction is still built in for support tickets, patient records, legal documents, internal notes, credentials, and structured identifiers.
Start free in a minute. Upgrade only when real traffic hits.
Common questions from developers.
Veil AI Firewall is a drop-in proxy and inspection API that secures prompts, responses, and MCP tool traffic. It includes the original Veil PII redaction flow plus prompt injection detection, output filtering, and MCP inspection.
Yes. PII redaction and response restoration are still built into the core proxy. The firewall layer expands Veil into prompt, response, and MCP runtime protection without removing the original privacy workflow.
On inline chat traffic, add headers like x-veil-input-policy: block or x-veil-output-policy: monitor. For standalone inspection, call /v1/firewall/input, /v1/firewall/output, or /v1/firewall/mcp.
Yes. Set the base_url parameter to your Veil endpoint and add your Veil API key in the headers. No other code changes needed. Works with the official SDKs and 41 named upstream providers.
It inspects MCP descriptors, tool calls, and tool results for tool poisoning, scope mismatches, destructive action patterns, prompt injection, suspicious links, and secret leakage.
Yes. Start on the free tier, use the live redaction demo above, point a staging client at Veil, or call the standalone firewall endpoints directly. You do not need to replace your provider or rewrite your app to try it.
