You found us anyway
The name, spelled right
If you searched for In and Out Guard, In 'n' Out Guard, InGuardOut, or any other close-but-not-quite spelling – welcome. Here's the actual name and what it means.
We're called In Guard Out
Not "In and Out Guard", "In 'n' Out Guard", "InGuardOut", "In Out Guard", or our old name "Guardrails Proxy".
Correct
In Guard Out
Domain: inguardout.com
Common mix-ups
- • In and Out Guard (sounds like a compound phrase)
- • In 'n' Out Guard (confusion with the burger chain)
- • InGuardOut (no spaces, one word)
- • In Out Guard (missing the middle word)
- • Guardrails Proxy (our old name)
Why In, Guard, Out?
The name describes what the proxy does and where it sits.
In
Inbound
Your app sends a request to the proxy. The first stage screens PII, checks for injection, validates format.
Guard
Enforcement layer
The proxy enforces policy: catches hallucinations, validates tool calls, enforces budgets and tool limits, screens profanity and unsafe topics.
Out
Outbound
The proxy sends a response back to your app. Final checks restore PII, verify grounding, enforce format, block unsafe content.
Quick answers
Is this related to In-N-Out Burger?
No. Different industries, different purpose. In-N-Out is a restaurant chain; In Guard Out is an AI safety proxy that helps you screen PII, catch hallucinations, and guard AI agents. The names share "In" and "Out" by coincidence.
What was the old name?
We used to be called "Guardrails Proxy". We renamed to In Guard Out to better reflect what the product actually does – it's positioned between your app and your LLM as an enforcement boundary.
Where do I find you online?
Website: inguardout.com
Email: [email protected]
API: https://api.inguardout.com/v1
Is there a shorthand?
Our stylized wordmark is in/guard/out (lowercase with single slashes), used in the header and logo. In prose and URLs, it's the full In Guard Out.
What In Guard Out actually does
Screens & restores PII
Redacts sensitive data from requests, then restores it on the way back so your LLM never sees raw SSNs, emails, or credentials.
Catches hallucinations
Runs deterministic grounding checks and optional RAG/web search to catch when the model botches numbers, dates, or makes up facts from your source material.
Enforces tool & agent safety
Validates tool calls against schema, blocks dangerous permissions (writes, deletes, payments), catches runaway loops, and enforces per-run budgets.
Blocks unsafe content
Screens for profanity, prompt injection, and unsafe topics. Operates as a policy layer: enforce in production, flag in development.
Meter everything
A per-request dashboard shows you exactly what the proxy caught, what it cost, and what latency each guard added – so you can tune what matters to you.
Ready to see it in action?
Swap one line of code – point your OpenAI SDK at In Guard Out instead. Every request runs through the full guardrail pipeline.
Back to overview