SUPERWISE Sentinel

The AI Gateway

Artificial Intelligence

Development

Security

Visit WebsiteSee on Product HuntTwitter ⧉

A few months into running Superwise Sentinel, here's my unfiltered take for anyone weighing it up.

The short version: it's a checkpoint between your application and the LLM that checks what goes in and what comes out. You set up guardrails for things like leaked secrets, PII and toxic language, then decide how each one responds. It can stop the request with a message you write, or quietly swap the sensitive bit for a placeholder and let it through. PII doesn't have to be all or nothing either; you pick the categories that matter to you.

The part I appreciate most is that inbound and outbound rules are configured independently. Protecting the model from your users and protecting your users from the model are two different problems, and Sentinel treats them that way.

Where it made a real difference:

• Real visibility instead of assumptions. The dashboard tracks interactions, latency, and blocked vs. remediated counts against the previous period, so it's obvious when a rule is pulling its weight and when it's just noise.

• A clear picture of spend. The FinOps view breaks token usage down in detail and helps optimize it. Turns out our costs were coming from places we didn't expect.

• A painless rollout. Starting with redaction rather than hard blocks meant we could enable protections without disrupting users, then tighten things as we learned.

Great work Team!

I've worked with teams adopting AI at different speeds, and governance often becomes an afterthought. What I like about Sentinel is that it helps organizations gain visibility and control without slowing down innovation or forcing major code changes. That balance between security and developer productivity is important. Great work by the team! 👏

AI calls can expose secrets and personal information. Sentinel routes them through a lightweight proxy that redacts sensitive data and logs activity—without code changes. Your prompts stay in your environment.

Setup takes about five minutes.

Been using Superwise Sentinel for a while now, so a few honest notes in case it helps someone deciding.

What it is, practically: it sits between your app and the model, and inspects both directions. You define guardrails — secrets detection, PII (you can scope it to specific categories rather than all-or-nothing), rudeness — and for each one pick an action: block with a custom message, or remediate by replacing the match with a placeholder. Separating input rules from output rules is the detail I'd flag, because the things you want to stop a user sending are not the same things you want to stop the model saying back.

What actually changed for us:

• We stopped guessing. The dashboard shows interactions, latency, and how much got blocked vs. remediated, with the prior period alongside it, so you can tell whether a rule you added is doing anything or just adding noise.

• Cost became visible. There's a separate FinOps view with token breakdown and consumption, plus token optimization. Our spend wasn't where we assumed it was.

• Rollout was low-drama. Redact-instead-of-block let us turn things on without breaking the product while we tuned the rules.

Not magic — you still have to think about which rules you want and where. But it replaced a pile of homegrown regex and a lot of hoping, and that's worth something.

Honestly the thing that matters most to me about Sentinel is that nobody has to touch their code. I've seen so many "just add our SDK" security tools die during rollout because someone's script or random agent never gets the integration and just... keeps doing its own thing. Here it's just an env var pointing at the proxy, so you actually get full coverage instead of hoping every engineer remembers to wire it in. Feels like the difference between a policy that exists on paper and one that's actually enforced.

Sentinel tackles a huge practical challenge in AI infrastructure: security and visibility can’t realistically be built into every single application or agent as usage explodes. Centralizing PII and Secret redaction, traffic observability, and policy controls in a lightweight gateway makes perfect sense.

Headroom is especially exciting. Tool outputs and retrieved context consume tokens fast, so optimizing that data pre-inference cuts costs without sacrificing context quality. The ability to tune compression based on content type and size adds great flexibility.

Packing security, visibility, and token efficiency into one unified layer feels super pragmatic for modern AI workflows.

Really nice work on Sentinel! 👏

Hey Hunters 👋
We're the team behind Superwise Sentinel — an AI Gateway that sits in front of your LLM traffic.

Point your apps at it (no code changes) and it redacts or blocks secrets, PII, and toxic content before they reach the model — with live visibility into everything. Your prompts and responses never leave your environment; only anonymous metadata reaches us.

Up and running in minutes via CLI, Docker, Helm, or a one-click Windows app. We'd love your feedback 🙏

Excited to see how Sentinel from Superwise can help healthcare enterprises protect their AI engagements for both business and quality of care applications.

Hey Product Hunters,

I'm John, co-founder and CPO at SUPERWISE. We've spent years in AI governance, and the pattern that finally made us build Sentinel was simple: the data that leaks into AI tools isn't leaking from some rogue app. It's people doing their jobs. A support rep pastes a customer record, an engineer pastes a stack trace with keys in it.

Sentinel runs on your own machine or network, close to where your data and apps already live. Passwords, API keys, health data and customer details are redacted before anything reaches the model, and every decision is logged so you can show what was caught.

Windows has a signed one-click installer. Mac and Linux install in a few minutes from the deployment guide. Every plan is free for 30 days, no credit card. If you'd rather see it first, the live demo runs a prompt through it with no install: superwise.ai/sentinel/demo

One question for you: what's the thing your team is most worried about pasting into an AI tool? I'll answer every comment today.