Since its launch, Kong AI Gateway has shipped as part of Kong API Gateway. Every AI capability we built (LLM routing, prompt guarding, semantic caching, token-based rate limiting) rode the same release train as the API gateway that powers some of the largest API platforms in the world. That coupling served customers well when AI traffic was a new workload finding its footing inside existing API infrastructure.

However, things have changed a lot over the last three years since we shipped AI Gateway. The agentic AI landscape moves on a clock of weeks, not quarters. New model providers appear, protocols like MCP and A2A evolve mid-flight, and the primitives teams need to govern agent traffic look less and less like the primitives they need to govern REST APIs. Meanwhile, the necessary API gateway release cadence is deliberately more conservative, because that's exactly what mission-critical API infrastructure demands. One codebase simply cannot honor both clocks.

So we split them. AI Gateway now has its own runtime, its own control plane, its own admin API, and its own version number — all starting at 2.0. Kong Gateway 3.x continues on its proven, stability-first quarterly cadence. AI Gateway 2.x moves at the speed of the AI ecosystem.

The version number signals the break cleanly: going forward, AI Gateway shouldn’t be considered a feature set inside Kong Gateway 3.x. It's a product with its own trajectory.

This change isn't just organizational. Because AI Gateway 2.0 stands on its own runtime and control plane, we were able to rebuild the entire experience around AI-native concepts:

• - **A dedicated runtime for AI traffic: **AI Gateway 2.0 runs as its own data plane, purpose-built for LLM, MCP, and agent workloads rather than inheriting the assumptions of an API-first runtime. It connects to a new AI Gateway control plane in Konnect, with its own analytics pipeline behind it.
• - **An admin API designed around AI workloads: **the new admin API treats Models, MCP Servers, and Agents as first-class entities with their own lifecycle, configuration, and policies. Policies, consumers, consumer groups, providers, and vaults round out the model. You configure the things you actually reason about when you run AI in production, instead of translating AI concepts into services and routes.
• - **A completely reimagined UX in Konnect: **new AI Gateway users in Konnect land directly in the AI Gateway 2.0 experience. Existing users will see their current setup untouched, alongside a new tab for AI Gateway 2.0 control planes. The new UI is organized around your models, your MCP servers, and your agents, with the policies and analytics that govern them one click away.
• - **A declarative path with kongctl: **AI Gateway 2.0 configuration is fully manageable through kongctl, so the new product fits into your existing GitOps workflows from day one. kongctl is also the migration vehicle: it converts existing decK configuration for AI Gateway plugins into native AI Gateway 2.0 config.
• - **Day-one provider coverage: **AI Gateway 2.0 ships with support for OpenAI and the major hyperscaler model platforms, along with support for Claude Code as a governed consumer of your AI infrastructure.

As great as the 2.0 user experience enhancements are, the main reason we did this is what comes next.

Decoupled from the API gateway release train, AI Gateway now ships on its own independent cadence, and that cadence will be fast. New providers, new policies, and new agentic primitives no longer wait for a quarterly platform release window. When the ecosystem moves, AI Gateway moves with it. You'll see the evidence quickly: AI Gateway 2.1 is already in flight and lands in August*. *You can expect some very exciting custom model cost management capabilities within the next two weeks, but more on this on release day.

This is the operating model the agentic era demands: a stable, battle-tested API gateway on one track, and an AI gateway that keeps pace with the fastest-moving infrastructure landscape in software on the other. One platform, one operating model for governance, and two release clocks that each match their workload.

We want to make sure it’s clear:** if you're running AI Gateway plugins on Kong Gateway today, nothing changes for you at launch**. Your current implementation keeps running, on the same telemetry pipeline, with the same compliance posture, in the same Konnect experience. AI Gateway plugins remain supported in Kong Gateway 3.14, as it is an LTS release. None of that changes with this new separation between 2.x and 3.x.

Two dates matter for planning:

• - **Kong Gateway 3.18** is when AI plugins in the API Gateway codebase become opt-in rather than bundled. We recommend completing your migration to AI Gateway 2.x before then, and we'll communicate that release date well in advance.
• - **Now** is when migration tooling is available. kongctl converts your existing decK configuration into AI Gateway 2.0 config, and your Kong account team can run a white-glove migration on your behalf from a decK dump of your current environment.
• - **End of July is when we plan to release Kong AI Gateway 2.x as generally available**

Start with the migration guide and the FAQ, which covers support timelines, LTS, certifications and compliance, and custom plugin questions. Your CSM is ready to help you scope a migration window.

AI Gateway 2.0 can be enabled in Konnect today. Once enabled, you can create an AI Gateway control plane and route your first LLM traffic in minutes. 

If you're interested in enabling AI Gateway 2.0 in your environment, reach out to your Kong CSM or main point of contact. We can enable the product and provide custom documentation during the private beta period.