Silker AISilker AI
Beta
Silker AI · Runtime application firewall · Shield, proxy, or SDK

Stop the attacks yourfirewall can't even see.

Silker AI is an AI-native security layer that blocks prompt injection, data leaks and app-logic attacks your WAF misses - in <10ms. Paste a script for any website, deploy a proxy for full runtime protection, or add one line of SDK code.

Prompt Injection Defense
Data Leak Prevention
<10ms Added Latency
Runs On Your Infra
Any site · paste in <head> · free
<script src="https://extra-api.silkerai.com/shield/embed.js"
data-site-id="site_..." defer></script>
Trust badge + link checks · no backend
Live · blocked in production
PROMPT INJECTIONInstruction override on /api/agent4ms
DATA LEAKAPI key in outbound response blocked2ms
IDORCross-tenant access on /orders/:id6ms
Free Security Scan

See your app through an attacker's eyes

Drop in your URL. We run a read-only scan of what's publicly visible - TLS, headers, exposed API docs, cookie settings - and send a plain-language report. No install, no exploit payloads.

Free passive scan after a quick sign-up. A full active pentest runs once you verify domain ownership - we never attack sites you don't own.

platform.silkerai.com
Silker AI Dashboard

One engine. Three ways to deploy.

The same detection engine, wherever your app lives. Pick the path with the least friction for your stack.

Recommended

Node.js SDK

One line in any Node app - Express, Next.js, Nest. Protects inbound + outbound traffic, including data leaks a WAF can’t see.

npm i @silker-ai/agent
Self-host

Docker Container

Run the proxy in front of any app, on any host. PHP, Java, Python, Go — no code changes required.

docker run ghcr.io/niceappspl/silker-proxy
Beta

Cloudflare Worker

Deploy a Worker on your Cloudflare zone. Any backend, zero code changes — we run inside Cloudflare.

wrangler deploy

Protects any stack

Silker works at the HTTP layer - it doesn't care what your backend is written in.

Node.js
Next.js
Python
Go
Java
.NET
PHP
Rust
Vercel, Netlify, Railway, any hostYour traffic never passes through our serversNo vendor lock-in
How It Works

Three steps to runtime security

Install, detect, monitor. Your traffic stays on your infrastructure - we only receive security events.

01

Install in minutes

One line in a Next.js app, a Cloudflare Worker, or a Docker container in front of any backend. No re-architecture.

npm i @silker-ai/agent
02

Detect locally, block instantly

Every request is inspected where your app runs. Threats - including prompt injection and data leaks - are blocked in real-time. Detection runs on your infra, not ours.

request → inspect → allow / 403
03

Monitor & fix

See blocked threats, data-leak alerts, and traffic on the dashboard. AI Copilot explains each issue and suggests the fix.

platform.silkerai.com
Features

Security a WAF doesn't give you

Silker is built for apps with AI in them. It blocks prompt injection, stops sensitive data from leaking, and covers the OWASP basics - all from one engine.

AI-Native Threat Detection

The threats unique to AI-generated and AI-powered apps - caught before they reach your model or database.

  • Prompt Injection & JailbreaksDetects instruction overrides, jailbreaks and extraction on your LLM routes.
  • OWASP Top 10SQLi, XSS, path traversal, file-upload abuse and more - included by default.
  • Rate Limiting & Auto IP BansStops abuse and brute-force with sliding-window limits and temporary bans - core steps in how to secure an API.
  • Scanner Trap NewHoneypot paths (/.env, /wp-login.php, /.git) bots probe first - hit one and we ban the IP instantly, before it finds a real hole.

Data Leak Prevention

The part a WAF can't see: sensitive data leaving your app. Silker inspects responses and outbound calls - why edge WAFs miss outbound data - then redacts or blocks leaks.

  • Secrets & API KeysCatches leaked keys, tokens, private keys and DB connection strings.
  • PII RedactionEmails, phone numbers, cards, SSN/PESEL - block, redact or just monitor.
  • On-demand Security ScansScan your live app for misconfigurations and exposed surface, with fix tips.

Monitoring & Alerts

One dashboard for every install - SDK, Worker or container. Deploy Silker as a reverse proxy in front of any backend and see blocked threats, data-leak alerts, traffic and geographic origins in real-time.

Real-time Logs
Traffic Analytics
Threat Map
AI Copilot
Live Activity Feed
GET /api/chat200 OK
POST /v1/chat (prompt injection)403 BLOCKED
GET /api/users (PII leak)REDACTED
POST /api/admin (SQLi)403 BLOCKED

Learn how runtime security works

Why Silker

How Silker compares

Every tool below covers part of the problem. Silker is the only one that connects the dots across your whole AI app - runtime, outbound data, AI routes and the database layer.

Cloudflare WAF
Great at the network layer, blind to your app logic and AI threats
Standalone WAF
Signature rules only - no understanding of AI or outbound leaks
LLM Firewall
Guards prompts but ignores your backend, DB and outbound data
SQLi / XSS / path traversal
Classic OWASP attacks on incoming requests
Cloudflare
WAF
LLM FW
Silker
Prompt injection & jailbreaks
Instruction overrides targeting your LLM routes
Cloudflare
WAF
LLM FW
Silker
Data leak prevention (outbound)
Secrets, API keys & PII leaving your app in responses
Cloudflare
WAF
LLM FW
Silker
Supabase / open database detection
Missing RLS, exposed anon key, unauthenticated routes
Cloudflare
WAF
LLM FW
Silker
App-context threats (IDOR, broken auth)
Access control flaws specific to your routes
Cloudflare
WAF
LLM FW
Silker
Rate limiting & API abuse
Sliding-window limits, auto IP ban
Cloudflare
WAF
LLM FW
Silker
No-code / one-command deploy
Works without touching app code (Worker or proxy)
Cloudflare
WAF
LLM FW
Silker
Works without a paid plan
Full protection on free Vercel / Railway / Render
Cloudflare
WAF
LLM FW
Silker
Real-time dashboard & AI threat analysis
Built for developers, not security rule writers
Cloudflare
WAF
LLM FW
Silker
Free security scan (no install)
Paste your URL, get a plain-language report
Cloudflare
WAF
LLM FW
Silker
Yes Partial No

Already on Cloudflare? Keep it - Silker adds the AI-native layer on top as a Worker or one line of code.

Pricing

Start free. Pay as you grow.

Detection runs on your own infrastructure - you pay for scale, advanced detection and the dashboard. Works with SDK, Cloudflare Worker or Docker.

Not sure which layer you need? See the best WAF options by use case.

Beta offer:50% off your first 3 months. Standard pricing after.
Starter
$0/mo

For side projects

  • 1 application
  • Up to 10,000 requests / mo
  • AI threat & OWASP blocking
  • 1 free security scan
  • Community support
Get Started Free
Most Popular
Pro
$19.50/mo$39

Beta: 50% off for first 3 months, then $39/mo

  • Up to 2 apps, 500k req / mo
  • AI Verdict & Copilot
  • Global Threat Map
  • Pentest & data leak detection
  • 30-day log retention
  • Email support
Get Pro
Business
$49.50/mo$99

Beta: 50% off for first 3 months, then $99/mo

  • Everything in Pro
  • Up to 5 apps, unlimited req / mo
  • Team seats & Google SSO
  • On-demand scans & Slack alerts
  • Priority support
Get Business
Enterprise
Custom

Self-host & regulated

  • On-prem / self-hosted
  • Unlimited apps & requests
  • SAML SSO, SIEM export
  • Custom SLA & dedicated support
Contact Sales

Prices in USD. Cancel anytime. Beta: 50% off your first 3 months, then standard pricing.

Documentation

Install in under 2 minutes

Pick one of three ways to deploy - same engine, same dashboard.

Node.js SDK Recommended

One line in any Node app - Express, Next.js, Nest. Inbound + outbound.

// Express / Node - zero-config (reads SILKER_API_KEY from env)
import { middleware } from '@silker-ai/agent'
app.use(middleware())

// Next.js (middleware.ts) - Edge-ready
import { nextMiddleware } from '@silker-ai/agent/next'
export const config = { matcher: '/api/:path*' }
export default nextMiddleware()

Cloudflare Worker Coming soon

Any backend behind Cloudflare. Runs on your zone.

# set SILKER_APP_ID + SILKER_TARGET in wrangler.toml
wrangler secret put SILKER_API_KEY
wrangler deploy

# Done - traffic is inspected at the edge.

Docker Container Self-host · any stack

Run the proxy in front of any app - PHP, Java, Python, Go. No Cloudflare required.

docker run -d \
  -p 8080:8080 \
  -e SILKER_TARGET=http://your-app:3000 \
  -e SILKER_API_KEY=sk_your_key \
  -e SILKER_APP_ID=your_app_id \
  ghcr.io/niceappspl/silker-proxy

# Point your traffic to :8080 - that's it.
The reality

Most AI-built apps ship with critical holes

Independent audits of apps built with Lovable, Bolt, Cursor, Replit and v0 keep finding the same thing: security is skipped by default. These are the numbers Silker exists to fix.

100%

of audited AI-built apps had at least one vulnerability

vibeAudit - 46 live apps, 2026
88–100%

shipped with the database left open - anyone could read every record

Independent vibe-coding audits, 2026
60%

of Lovable apps exposed an API key directly in the browser

vibeAudit research, 2026
1.5M

auth tokens + 35k emails leaked from one app within days of launch

Moltbook breach, 2026

These aren't edge cases - they're the default output of AI coding tools. Silker is the runtime layer that blocks these attacks live, instead of just emailing you a report after the breach.

Onboarding our first design partners - early access open
// measured in production

Numbers, not promises.

Verified against Silker's own proxy instrumentation and coverage matrix — not industry estimates.

<10ms

proxy latency overhead p99 — real-time inspection, no user-visible lag

Measured · reverse proxy mode
0

lines of application code changed — drop in a Docker proxy or paste a script tag

docker run · <script> · npm i
~2 min

to full runtime protection — Docker reverse proxy, zero code changes

docker run · paste & go · npm i
100%

of requests inspected — every inbound payload and every outbound response, in real time

Complete bidirectional inspection by design

These are Silker's own production metrics — not vendor benchmarks. See how we measure →

Ship secure AI apps

Install in one line of code. Catch prompt injection and data leaks before your users do - with less than 10ms added latency.