See everything your AI sends.
Control everything it costs.
The only OpenAI-compatible proxy that does DLP, billing, and per-project governance in a single stateless hop. Redact PII, block prompt injections, enforce spend limits, and smart-route to the cheapest provider—before a single token leaves your system.
Native SDK (pip install aisg) or OpenAI SDK compatible · No credit card required · 600+ models
Not sure if your prompts leak data? Test for free →
What your LLM provider actually receives
Without AISG
"prompt": "Summarize this contract for John Smith (john@company.com). His SSN is 123-45-6789 and the deal is worth $2.4M with Acme Corp."
}
With AISG
"prompt": "Summarize this contract for [PERSON] ([EMAIL]). His [US_SSN] and the deal is worth $2.4M with [COMPANY]."
}
Names, emails, SSNs, and company details redacted. Business context passes through untouched.
Instantly see what PII your prompts leak, what each call costs, which provider is cheapest—with just 2 lines of code.
Works with every major provider
Is your prompt leaking data?
Paste any AI prompt below. Our scanner detects 28+ sensitive data types instantly.
You need this if…
Your app sends user data to GPT, Claude, or any LLM
You don’t know what developers are putting into prompts
You need to enforce PII, PCI, or HIPAA compliance on AI calls
You want cost control across multiple AI providers
You run autonomous agents that could get stuck in infinite retry loops
You need real-time alerts when PII leaks, injections, or budget breaches happen
You need EU AI Act Article 12 compliance logging before August 2026
You’re paying for identical LLM prompts that could be served from cache
You need a stateless proxy that never stores prompts or trains on your data
You need project-level versioned DLP Policies & custom regex IP protection
Two lines of code. Full governance.
Use the aisg SDK for typed metadata, or keep the OpenAI SDK — switching takes 30 seconds either way.
const openai = new OpenAI({
apiKey: 'OPENAI_KEY'
});const openai = new OpenAI({
apiKey: 'os_hub_live_abc123',
baseURL: 'https://api.aisecuritygateway.ai/v1'
});~50ms
PII scan latency
28+
PII entity types
600+
Models, 8 providers
The Request Journey
Your App
Sends the request
DLP / AI Firewall
Scans & redacts PII
Smart Router / Budget
Cheapest provider & spend cap
Provider Routing
BYOK / Managed / Failover
AI Provider
Only sees clean, governed data
Why AI Security Gateway
The only OpenAI-compatible proxy with
DLP, billing, and per-project governance built in.
Cloudflare AI Gateway does caching and rate limiting — zero DLP. Portkey does routing — no PII scanning. We do both, in a single stateless hop.
| Capability | Cloudflare AI Gateway | Portkey AI Gateway | Helicone Observability | AISG Governance |
|---|---|---|---|---|
| PII redaction (28+ entities) | — | 3rd-party | — | |
| Vision / image OCR scanning | — | — | — | |
| Prompt injection blocking | — | 3rd-party | Partial | |
| Hard budget enforcement (402) | — | Partial | — | |
| Per-project DLP policies | — | — | — | |
| Custom regex IP protection | — | Partial | — | |
| Managed credits / wallet | — | — | — | |
| Smart cost routing | — | — | ||
| BYOK at 0% markup | — | |||
| Multi-provider failover | — | |||
| Recursive loop protection | — | — | — | |
| Webhook security alerts | — | Partial | ||
| EU AI Act compliance logging | — | — | — | |
| Semantic prompt caching | ||||
| Deep observability / tracing | — | Metadata | ||
| Open source | — |
Why this matters: Palo Alto Networks acquired Portkey in May 2026 to get the routing half of this equation. But Portkey has zero native DLP. Cloudflare AI Gateway does caching and rate limiting but no PII scanning. Helicone does observability but no active intervention. AI Security Gateway (AISG) is the only proxy that does DLP + billing + multi-project governance in a single stateless hop.
No credit card required · Integrate in 2 minutes
The obvious question
“Why not just…”
Every team asks. Here’s why combining two tools doesn’t equal one governance layer.
Why not just use Cloudflare AI Gateway + a separate DLP tool?
Because that’s two systems, two failure points, and zero coordination between routing decisions and security policies. AISG runs DLP inside the proxy path — every request is scanned before it leaves your network, with per-project policies and budget enforcement, in a single API call. No sidecar. No second hop.
Why not just use Portkey for routing and add PII scanning later?
Portkey routes requests but has zero native DLP. Palo Alto acquired them for $356M and still doesn’t have inline PII redaction. Bolting DLP onto a routing proxy after the fact means prompts hit an external scanner, adding latency and a second trust boundary. AISG scans and routes in the same stateless hop.
Why not build our own DLP + proxy internally?
You absolutely can — our open-source version is Apache 2.0. But most teams underestimate the scope: entity detection across 28+ PII types, prompt injection heuristics, per-project versioned policies, budget enforcement with HTTP 402, multi-provider failover, and keeping pricing data current. That’s 6–12 months of engineering. AISG gives you all of it in two lines of code.
The one-liner: “The only OpenAI-compatible proxy that does DLP, billing, and per-project governance in a single stateless hop.”
No credit card required · Integrate in 2 minutes
What you get on every request.
28+ PII entity redaction
Names, SSNs, credit cards, emails, API keys — scanned and redacted on every request, including images via OCR.
Project-level DLP policies
Define versioned data-loss-prevention policies per project. Different teams, different rules — all enforced automatically at the proxy layer.
Custom regex patterns
Write your own regex rules to catch proprietary terms, internal project names, or domain-specific data that standard PII detectors miss.
Prompt injection blocking
Heuristic jailbreak detection that blocks malicious prompts before they reach any provider.
Hard budget enforcement
Per-key spend caps that return HTTP 402 when exceeded. No runaway agents, no surprise bills.
Recursive loop protection
Detects and kills agent retry loops automatically. Fingerprint-based detection blocks repetitive request patterns before they drain your budget.
Semantic prompt caching
Identical DLP-cleaned prompts are served from cache — zero LLM cost, sub-millisecond latency. Shared across all proxy instances via a low-latency distributed cache.
Smart cost routing
Automatically routes each request to the cheapest available provider across your configured models.
Webhook notifications
Real-time HMAC-signed alerts to your Slack, PagerDuty, or SIEM when PII is blocked, budgets are hit, or loops are detected.
EU AI Act compliance logging
Tamper-evident, hash-chained audit records on every request. Export as JSONL. Ready for Article 12 enforcement (August 2026).
BYOK at 0% markup
Bring your own API keys. Pay providers directly with zero AISG markup on Pro. Full governance still applies.
Open source
Self-host the entire stack in your own cloud. Apache 2.0 licensed. No vendor lock-in.
No credit card required · Integrate in 2 minutes
Built for zero-trust AI usage.
Your prompts are governed in transit and never retained. We don’t store what you send, and we never train on your data.
Stateless proxy
Prompts pass through and are never written to disk or logged. Only metadata (cost, latency, entity counts) and SHA-256 fingerprints (for EU AI Act compliance) are recorded — never content.
Never train on your data
Your requests are yours. We do not use prompt content for model training, analytics, or any secondary purpose.
AES-256-GCM encryption
BYOK API keys are encrypted at rest with authenticated encryption. Keys are decrypted only at request time in memory.
Self-host in your VPC
Deploy the full open-source stack on your own infrastructure. Apache 2.0 licensed. No phone-home, no telemetry.
Simple pricing. No surprises.
Three plans. All include the full governance layer across 600+ models.
Managed Credits
For prototypes & teams who want zero provider setup
- 1M free credits — no credit card
- We provide AI access (no provider signups)
- Smart Router picks cheapest provider
- Full AI Firewall: 28+ PII types
- 25% open-source / 30% closed markup
Starter BYOK
For solo devs & single projects
- 3 projects — bring your own keys
- 0% Hub markup on all calls
- All premium models across providers
- Full AI Firewall: 28+ PII types
- AES-256 encrypted key storage
Pro BYOK
For teams and production
- Unlimited projects — bring your own keys
- Everything in Starter, plus:
- Priority support
Full pricing details, media generation tiers, and plan comparison →
FAQ
Frequently Asked Questions
It’s a security layer that sits between your app and AI companies like OpenAI, Groq, or Mistral. When your app sends a message to AI, AISG checks it first — removes any sensitive data (like emails, passwords, or credit card numbers), selects the most cost-efficient available provider, and then forwards the cleaned message. Think of it as a security guard and cost-optimization engine combined.
No. We only log metadata (like "1 email address was blocked") — we never store the actual content of your messages or the AI’s responses. Your data passes through our security layer and is forwarded to the AI provider. Nothing is saved on our servers.
If your app uses the OpenAI SDK (the most popular AI library), you only need to change two lines: the API key and the base URL. Everything else — your models, your prompts, your response handling — stays exactly the same.
BYOK stands for "Bring Your Own Key." It means you already have an account (and API key) with a provider like OpenAI, Google Gemini, xAI, Groq, or Together.ai. On the Pro plan, you save those keys in AISG, and we use them to make AI calls on your behalf — so you pay the provider directly at their regular price with zero markup. BYOK also unlocks premium models like GPT-4o and Gemini Pro.
That’s perfectly fine! The "Managed Credits" plan is designed exactly for this. Add credits to your AISG wallet (10M Credits / $10 minimum), and we handle everything — we use our own provider accounts to process your requests.
No. Every request is protected from your very first API call. AISG applies a "Maximum Protection" default policy that scans for all 28 entity types (emails, credit cards, SSNs, API keys, prompt injection attacks, and more) and redacts any matches before the AI model sees them.
The security check adds less than 50 milliseconds for text — that’s faster than a blink. For images, it takes about 0.5–1 second depending on image size. Every response includes timing headers so you can verify this yourself.
Yes. Pro subscriptions are managed through Stripe — cancel with one click anytime. Wallet balances never expire, so any prepaid credits stay in your account.
More questions? Check our full documentation for in-depth guides on AISG Credits, providers, DLP policies, and more.
No credit card required · Integrate in 2 minutes
Stay in the loop
New models, security features, and cost-saving updates. No spam.
As Featured On
