What is a personalized AI gateway?
AI gateways sit between users and AI to govern and scale AI applications. Personalized AI gateways extend this pattern, sitting between users and internet applications to hydrate AI calls with user-owned runtime context.
This simple extension unlocks a new type of internet infrastructure – one where every application can adapt to us based on our complete context, not just what we do in any single app.
AI gateways are increasingly popular in enterprise
In September 2023, Cloudflare launched their AI Gateway to make "AI applications more observable, reliable, and scalable."
AI Gateway sits between your application and the AI APIs that your application makes requests to (like OpenAI) – so that we can cache responses, limit and retry requests, and provide analytics to help you monitor and track usage.
Databricks followed suit in September 2024, launching their own AI gateway with additional governance features.
From web gateways to AI gateways
These gateways follow Secure Web Gateway (SWG) patterns, which sit between users and the internet. Palo Alto Networks summarizes:
A SWG is a cloud-delivered network security technology that filters internet traffic and enforces regulatory policy compliance. SWGs inspect traffic from client devices aiming to connect to the internet, they authenticate users and examine web requests blocking those that violate policies or those that pose a security threat.
SWGs are typically deployed by enterprises to protect employee web traffic. But this architectural pattern – a trusted intermediary that governs interactions between users and services – extends naturally to personalization.
Today, applications wishing to personalize experiences typically:
- Instrument user behavior to collect local context
- Activate this context with AI models
But for the age of AI, this fragmented context is proving insufficient. Every app builds its own data silos, yet none have the complete picture needed for truly adaptive experiences.
Here's how this personalized AI gateways look in practice:

The trouble with integration protocols
Integration protocols like Model Context Protocol (MCP) are rising to fix this fragmentation. MCP enables users to bring their context to AI clients, but their privacy properties are poor.
MCP follows a "context to compute" pattern, where users send their context to possibly untrusted third party AI clients. This creates several problems:
- Context hunger: MCP clients may consume much more context than needed to answer simple questions like:
- When am I free next week?
- Where should I go for dinner?
- What type of sofa might I like?
- Privacy violations: This may violate CCPA and GDPR's data minimization principle. CCPA requires that data use be "reasonably necessary and proportionate" to the purpose. But MCP clients can access broad interfaces like the following without being bound to any purpose limitation:
search_all_email()
list_recent_transactions()
- User burden: Users must understand and consent to complex server interfaces, creating cognitive overhead that leads to consent fatigue.
The case for compute to context
A personalized AI gateway can fix this by moving compute to context. Instead of sending context to compute, applications call a user's AI agent to get inference based on permissioned context.
Here's how it works:
// Define what context you need
const contextQuery = {
"recent_travel": {
"select": ["object_name", "object_style"],
"from": "personalTimeline",
"where": [
{"field": "object_subcategory", "op": "=", "value": "HOTELS_ACCOMMODATIONS"}
],
"limit": 10,
"orderBy": [{"field": "originalTimestamp", "dir": "desc"}]
}
};
// Call the gateway
const response = await fetch("https://ai.crosshatch.io/", {
headers: {
"Authorization": `Bearer ${userToken}`,
"Crosshatch-Provider": "gpt-4o",
"Crosshatch-Replace": JSON.stringify(contextQuery)
},
body: JSON.stringify({
messages: [{
role: "user",
content: "Based on their travel history {recent_travel}, where might they like to stay in Paris?"
}]
})
});
The gateway may provide all the features of enterprise AI gateways:
- Logging
- Caching
- Rate limits
- Retries
- Support for all top model providers
But adds critical new capabilities:
- Runtime context hydration
- User-governed sharing policies
- Fine-grained access controls
- Automatic context synchronization
The shift to personalized AI gateways
Crosshatch is the Internet's first personalized AI gateway.
We created Crosshatch because we believe that every application on the internet should adapt to us based on our complete context.
The modern data stack, which powered the last generation of personalized applications, has become bloated, unwieldy and expensive while failing to deliver the promise of adaptive applications.
Personalized AI gateways align with the new patterns of MCP – enabling users to port their context into third party applications – but do so in a way that's privacy compliant and easy for consumers to understand.
Want to see how it works?
- Try our playground
- Check out our docs
- Create a developer account
Join us in building an internet that adapts to each user.