Inline defense against AI attackers
The gate AI attackers can't see through.
hollowgate sits between your firewall and your apps. It spots the AI agents and bots probing you, catches them with built-in deception that flags attackers who go looking where real users never do, and keeps them out at machine speed, while people sail through.
Runs on your hardware or any cloud. Starts in watch mode. Strictly defensive: it never touches anyone else's systems.
Why now
Attackers stopped typing.
Reconnaissance, credential stuffing and exploit attempts are increasingly run by AI agents that work around the clock, adapt in seconds and look a lot like people. Tools built for human-speed attacks and human-speed analysts fall behind.
They adapt faster than rules
An agent that gets a 403 tries a new path, a new header, a new pace. Static signatures and rate limits get probed and worked around.
They pass basic bot checks
Real browsers, residential IPs and solved challenges mean "is this a bot" is the wrong question. The useful question is "is this behaving like an attacker".
Your team works at human speed
By the time an alert is triaged, the agent has mapped the app. Defense has to decide and act in milliseconds, then explain itself to the analyst.
Vulnerability shield
Protected the day a CVE drops, not the day you patch.
New vulnerabilities are being exploited within hours of disclosure, often by automated agents scanning the whole internet. Patching takes days or weeks. hollowgate closes that window for you.
Knows what you run
hollowgate learns the apps and frameworks behind it, so it knows which new vulnerabilities actually matter to you.
Watches every day
Newly published and actively exploited vulnerabilities are checked against your apps daily, with the ones attackers are already using first.
Tightens protection until you patch
Affected apps get stronger protection automatically. When you patch, it steps back on its own.
Tells your team what to fix first
Your analysts see which apps are exposed, what is being exploited in the wild, and what is covered while they patch.
How it works
Watch every session. Trap the curious. Act in milliseconds.
hollowgate is a reverse proxy you put in front of your apps. Decisions are made on the node itself, with no internet round trip, so they don't wait on anyone.
Sessions, not single requests
It looks at how each visitor behaves over time and learns what normal looks like for each of your apps.
A verdict, explained
Each session gets a verdict with plain-language reasons your analysts can check. Partner agents and your own scanners can be allowlisted.
Deception built in
Built-in deception that flags attackers who go looking where real users never do. Your customers never notice it.
Graduated and self-expiring
Throttle first, block when the evidence is strong. Every action expires on its own, so a false positive heals itself.
Normal sessions pass straight through. Real users never see a challenge page.
Suspicious sessions slow down. Attackers lose time; people barely notice. It lifts on its own.
Only when the evidence is strong. Blocks lift on their own unless an analyst confirms them.
The AI judge never sits in the path of your normal traffic. Use your own Anthropic or OpenRouter key, or run a local model in an air-gapped site.
Deception, built in
A gate that looks open, and is hollow.
Agents are curious by design. hollowgate turns that curiosity against them, with deception that flags attackers who go looking where real users never do.
Invisible to your users
Your customers never see it and never trip it, so it adds no friction.
Evidence, not guesses
A session that goes looking has shown intent. That makes blocks fast and false positives rare.
Built for AI agents
Designed around the way LLM-driven attackers explore an app, and updated as their techniques change.
Where it fits
Built for the attacker that changed.
Keep your WAF and your CDN. hollowgate covers the gap between them: adaptive, automated attackers that look legitimate request by request.
| WAF | Bot management | Deception platforms | hollowgate | |
|---|---|---|---|---|
| Main question | Is this request malicious? | Is this a bot? | Did someone touch a lure inside the network? | Is this session behaving like an attacker, human or AI? |
| Sees | One request at a time | Client fingerprints, challenges | Internal hosts and credentials | Whole sessions in front of your apps |
| Catches AI agents that pass bot checks | Rarely | Partly | After they're inside | Yes: behavior plus traps at the edge |
| Covers new CVEs before you patch | When a rule update ships | Rarely | No | Yes, automatically, for the apps you run |
| Acts on its own | Block by rule | Challenge or block | Alert | Throttle or block, self-expiring |
| Where it runs | Varies | Usually vendor cloud | Your network | Your hardware, any cloud, or air-gapped |
| Hurts real users | On false positives | CAPTCHAs and friction | No | No challenges; mistakes expire on their own |
These are typical category traits, not claims about any specific vendor. Many teams run hollowgate alongside their existing WAF.
For the SOC
Every decision, with its reasons.
A console your analysts will actually keep open: what was decided, why, and a one-click way to lift, confirm or block.
Live session timelines
Every request in a session and the reasons behind each decision, in one view.
Watch mode first
New installs record what they would have done without acting. Turn on enforcement when the numbers look right to you.
SIEM-ready events
Structured JSON events for Splunk, Sentinel, Elastic or anything that reads a log file.
Alerts by email
Bring your own SMTP server. Trap trips and blocks reach the on-call inbox in seconds.
Your AI, your keys
Plug in an Anthropic or OpenRouter API key from the settings page. Only redacted samples are ever sent, and only to the provider you chose.
Analyst in the loop
Lift a block, confirm it, or block a session by hand. Every action is audited.
Deploy anywhere
Your hardware. Your cloud. Your rules.
hollowgate ships as a package you run, sized from a single small VM to a fleet behind your load balancer. Licenses verify offline, so it works the same in an air-gapped data center.
Docker
One image in front of one app or many. The quickest way to try it.
Kubernetes
As an ingress or a sidecar, scaled with the rest of your cluster.
VM appliance
A hardened virtual appliance for VMware, Hyper-V, AWS, Azure or GCP.
Your servers
Any Linux host next to your firewall. No GPU needed on the data path.
Self-hosted
Everything on your side. Nothing leaves your network unless you turn on an AI provider.
Hybrid
Nodes on your side, AI judgments through your own provider key. The most common setup.
Air-gapped
A local open-weight model, or local rules only. Offline license, offline updates.
Principles
A security tool you don't have to defend.
Anything inline becomes a target. hollowgate is built to be hard to attack and safe to run.
Strictly defensive
hollowgate only acts on your own infrastructure. It never connects back to, probes or interferes with anyone else's systems.
Hardened by default
Locked down out of the box. Secrets stay out of logs and are never shown again once saved.
Never a single point of failure
Every action self-expires. A license problem can pause paid extras, but it never blocks your traffic and never turns protection off.
Your data stays yours
Decisions happen on the node. Only redacted samples go to the AI provider you chose, and only if you chose one.
Pricing
Start free. Pay when it earns its place.
Every edition includes full detection, traps and enforcement. Paid editions add the vulnerability shield, the AI judge, alerting, SIEM export and scale.
Free1 node, for one app or a lab
- Attacker detection and automatic response
- Built-in deception
- Console and event log
- Watch and enforce modes
Per node, billed yearlyUp to 3 nodes included · pricing on request
- Everything in Community
- Vulnerability shield: daily CVE watch with automatic protection
- AI judge with Anthropic or OpenRouter
- Self-learning baseline per route
- Email alerts and SIEM export
- Email support
CustomFleets, regulated and air-gapped sites
- Everything in Pro
- Multi-node with shared sessions
- Fleet-wide learning
- SSO for the console
- Air-gapped AI judge
- Priority support
FAQ
Questions security teams ask first.
Does hollowgate attack the attacker?
No. It never hacks back, scans or touches anyone else's systems. Everything it does happens on your own infrastructure: watching sessions, slowing them down or refusing them.
What if it blocks a real customer?
It starts in watch mode so you can see its decisions before it acts. In enforce mode it throttles before it blocks, blocks only when the evidence is strong, and every action expires on its own. Analysts can lift any block in one click.
Does our traffic leave our network?
Not by default. Decisions are made on the node. If you connect an AI provider, only redacted samples are sent, through your own API key.
Will it slow our site down?
Scoring is lightweight work on the node and normal sessions pass straight through. The AI judge is never in the path of ordinary traffic.
What happens if our license expires?
You get reminders from 30 days out and a grace period after expiry. After that, paid extras such as the AI judge and alerting pause until you renew. Detection, traps and enforcement keep running, and traffic is never blocked because of a license.
Does it replace patching?
No. It buys you time. When a new vulnerability affects one of your apps, hollowgate protects it automatically from the day it's disclosed, so your team can patch on a sane schedule instead of in a panic.
Do we need GPUs?
No. The data path runs on ordinary CPUs. You only need a GPU if you want to run your own AI model in an air-gapped site.
Book a demo
See it catch an agent in your own traffic.
We'll set hollowgate up in watch mode in front of a test app or a staging copy of yours, and walk through what it finds.
Or email [email protected]