Kgateway is an open source implementation of the Kubernetes Gateway API that unifies ingress, API gateway, service mesh, and AI gateway capabilities in a singular modular control plane. Built for performance and flexibility, it secures and manages traffic across legacy, cloud native, and AI-driven workloads in any environment.
We’re excited to announce the release of kgateway v2.1., a release packed with exciting new features and improvements. Here are a few select updates the kgateway team would like to highlight!
🌟 What’s new in kgateway 2.1?
Agentgateway integration
This release marks a major milestone — it’s the first version to integrate the open source project agentgateway! Agentgateway is a highly available, highly scalable data plane that provides AI connectivity for LLMs, MCP tools, AI agents, and inference workloads. As part of this evolution, we’re beginning the deprecation of the Envoy-based AI Gateway and the Envoy-based Inference Extension, since all related functionality is now implemented natively through agentgateway. You can still continue to use Envoy-based Gateways for API Gateway use cases.
For this release, agentgateway support is in beta. If you’re trying out the agentgateway GatewayClass, we recommend following the beta release feed to stay up to date with improvements, bug fixes, and breaking changes as the implementation is refined.
To get started with agentgateway, you simply install kgateway with the following Helm values:
agentgateway: enabled: true
Then you create a Gateway with the agentgateway GatewayClass as shown here:
kubectl apply -f- <<EOF kind: Gateway apiVersion: gateway.networking.k8s.io/v1 metadata: name: agentgateway namespace: kgateway-system labels: app: agentgateway spec: gatewayClassName: agentgateway listeners: - protocol: HTTP port: 8080 name: http allowedRoutes: namespaces: from: All EOF
You are now ready to try out agentgateway. Check out the agentgateway guides to learn how to route traffic to an LLM provider, MCP tool server, or AI agent.
Check out the accompanying examples demonstrating how agentgateway connects to MCP servers and A2A agents:
K8s GW API 1.3.0 and Inference Extension 1.0.0
Kgateway is now fully conformant with the Kubernetes Gateway API version 1.3.0 and Inference Extension version 1.0.0. For details, check out the conformance test reports:
Global policy attachment
By default, you must attach policies to resources that are in the same namespace. Now, you can enable a feature to create a global namespace for policies. These global policies can attach to resources in any namespace in your cluster through label selectors. For more information, see the Global policy attachment docs.
Deep merging for extauth and extproc policies
You can now apply deep merging for extAuth and extProc policies. In addition, you can use the kgateway.dev/policy-weight annotation to determine the priority in which multiple extAuth and extProc policies are merged. For more information, see Policy priority during merging.
Additional proxy pod template customization
kgateway now has more options to customize the gateway proxies’ default pod template, including configuration for nodeSelectors, affinity, tolerations, topologySpreadConstraints, and externalTrafficPolicy.
For more information, see Customize the gateway. To find all the values that you can change, see the PodTemplate reference in the GatewayParameters API.
Horizontal Pod Autoscaling
You can bring your own Horizontal Pod Autoscaler (HPA) plug-in to kgateway. This way, you can automatically scale kgateway control and data plane pods up and down based on thresholds like CPU or memory consumption. See Horizontal Pod Autoscaling (HPA) for more information.
Dynamic Forward Proxy
You can now configure the gateway proxy to use a Dynamic Forward Proxy (DFP) filter. This filter allows the proxy to act as a generic HTTP(S) forward proxy without the need to preconfigure all possible upstream hosts. Instead, the DFP dynamically resolves the upstream host at request time by using DNS. Check out Dynamic Forward Proxy (DFP) for more information.
Session Affinity
You can now configure different types of session affinity for your Envoy-based gateway proxies:
- Change the default loadbalancing algorithm: By default, incoming requests are forwarded to the instance with the least requests. You can change this behavior and instead use a round robin or random algorithm to forward the request to a backend service.
- Consistent hashing: Set up soft session affinity between a client and a backend service by using consistent hashing algorithms.
- Session persistence: Set up “strong” session affinity or sticky sessions to ensure that traffic from a client is always routed to the same backend instance for the duration of a session.
Enhanced retries and timeout capabilities
Retries and timeout capabilities have been enhanced for your Envoy-based gateway proxies. Check out the following guides for more information:
Passive health checks with outlier detection
You can now configure passive health checks and remove unhealthy hosts from the load balancing pool with an outlier detection policy. An outlier detection policy sets up several conditions, such as retries and ejection percentages, that kgateway uses to determine if a service is unhealthy. When an unhealthy service is detected, the outlier detection policy defines how the service is removed from the pool of healthy destinations to send traffic to. For more information, see Outlier detection.
New kgateway operations dashboard
When you install the OTel stack, you can now leverage the new kgateway operations dashboard for Grafana. This dashboard shows important metrics at a glance, such as the translation and reconciliation time, total number of operations, the number of resources in your cluster, and latency.
Check out this demo of some of kgateway’s new features in global policy attachment, HPA, monitoring, and observability:
🗑️ Deprecated or removed features
AI Gateway and Inference Extension support for Envoy-based gateway proxies is deprecated and is planned to be removed in version 2.2. If you want to use AI capabilities, use an agentgateway proxy instead. To learn more about why we think that agentgateway is better suited as a gateway for agentic AI and MCP workloads, check out this blog.
Release notes
Check out the full details of the kgateway v2.1 release in our release notes.
Availability
Ready to get started? Download the latest release on GitHub. Then, check out our getting started guide to install kgateway.
Thanks to our contributors!
The kgateway v2.1 release includes 602 contributions from 42 contributors. Check out the project dev stats for the v2.1 release cycle. We would like to extend our sincere gratitude to all the contributors!
This release also marks kgateway’s first participation in the LFX Mentorship Program! Our mentees have contributed across a range of initiatives across the project, including automated scale tests for kgateway, improving ecosystem integrations documentation, OpenTelemetry observability for AI Extensions, and observability improvements for agentgateway. Aryan Parashar, one of our current Term 3 mentees, also co-authored this blog!
Get Involved
The simplest way to get involved with kgateway is by joining our Slack and community meetings.
Thank you for your continued feedback and support!