Release notes

Review the release notes for agentgateway.

v2.2.0

Version 2.2 introduces major breaking changes for agentgateway, including new dedicated APIs, a split Helm installation, and documentation moved to agentgateway.dev.

ℹ️
If you are currently running agentgateway with kgateway on version 2.1, a new installation is recommended.

If you choose to upgrade to version 2.2, review the following considerations.

  • Upgrading the control plane automatically restarts any agentgateway data plane proxies.
  • The custom resource APIs, default namespace, controller, and other settings are updated.

Continue reviewing the release notes to understand the changes from the previous version.

🔥 Breaking changes

Note that the previous 2.1 version of agentgateway on Kubernetes used the kgateway project as the control plane, as well as the kgateway APIs such as TrafficPolicy. In 2.2, these old kgateway APIs are no longer supported for agentgateway.

ℹ️

Dedicated agentgateway APIs and installation

Agentgateway now has dedicated APIs and a separate installation from kgateway:

  • New APIs in the agentgateway.dev API group
  • New AgentgatewayPolicy API to replace TrafficPolicy for agentgateway policy configurations
  • New AgentgatewayParameters API to replace GatewayParameters for agentgateway proxy configurations
  • Split Helm installation with dedicated charts for agentgateway

Key changes include:

  • Policies are now configured through AgentgatewayPolicy instead of TrafficPolicy
  • DirectResponse for agentgateway is now only configurable through AgentgatewayPolicy instead of the separate DirectResponse CRD
  • Agentgateway can no longer be configured with GatewayParameters, only with AgentgatewayParameters
  • The controller name changed from kgateway.dev/agentgateway to agentgateway.dev/agentgateway
  • AgentgatewayParameters rawConfig breaking change to allow configuring binds and other settings in config.yaml outside of its config section
  • The default namespace for agentgateway is now agentgateway-system instead of kgateway-system
ℹ️
For a detailed comparison of agentgateway vs kgateway resources, including GatewayClass, controller names, Helm chart locations, and CRDs, see the kgateway v2.2 release blog.

Feature gate for experimental Gateway API features

The KGW_ENABLE_EXPERIMENTAL_GATEWAY_API_FEATURES environment variable gates experimental Gateway API features and APIs. This setting defaults to false and must be explicitly enabled to use experimental features such as XListenerSet, Route SessionPersistence, HTTPCORSFilter, and HTTPRouteRetry.

To enable these features, use a Helm values file or the --set flag during installation:

controller:
  extraEnv:
    KGW_ENABLE_EXPERIMENTAL_GATEWAY_API_FEATURES: "true"

Or use the Helm flag: --set controller.extraEnv.KGW_ENABLE_EXPERIMENTAL_GATEWAY_API_FEATURES=true

ExtAuth fail closed

Agentgateway ExtAuth policies now fail closed when the backendRef to the auth server is invalid. Previously, invalid backend references might have allowed requests through. Update your ExtAuth policies to ensure backend references are valid before upgrading.

AI prompt guard API alignment

The AI prompt guard API is updated to align with other enums. The values changed from MASK to Mask and REJECT to Reject. These changes are enforced by CEL validation in the API. Update any existing prompt guard configurations accordingly.

🌟 New features

Performance and infrastructure

Performance improvements: The agentgateway control plane was refactored, improving performance by up to 25x.

Multi-arch controller image support: Agentgateway now supports multi-architecture controller images.

Infrastructure options:

  • Gateway.spec.addresses support for configuring load balancer IP addresses
  • PodDisruptionBudget and HorizontalPodAutoscaler options via AgentgatewayParameters
  • Event reporting for agentgateway gateways that indicates when a gateway has NACKed an update

AI and LLM support

Model aliases: Added modelAliases support to AgentgatewayPolicy to allow friendly model name aliases for your AI backends (for example, “fast” can map to “gpt-3.5-turbo”).

Provider support:

  • Added support for Azure OpenAI backends
  • Added support for multiple AI backend route types including OpenAI Responses API, Anthropic token counting, and prompt caching configuration for Bedrock (enabling up to 90% cost reduction)
  • Path-based API format routing (completions, messages, models, passthrough) to enable a single backend to support multiple LLM API formats based on request URL

Canadian Social Insurance Number prompt guards: Added support for detecting and guarding Canadian Social Insurance Numbers in prompts.

MCP support

MCP authentication: MCP authentication enables OAuth 2.0 protection for MCP servers, helping to implement the MCP Authorization specification. Agentgateway can act as a resource server, validating JWT tokens and exposing protected resource metadata.

Stateful/stateless session routing: You can now configure the MCP session behavior for requests to be Stateful or Stateless on the AgentgatewayBackend. Behavior defaults to Stateful if not set.

Multi-network support: Added support for cross-network workload discovery and routing in ambient mode.

Authentication and security

  • Basic auth, API key auth, and JWT auth: Agentgateway proxies now support basic auth, API key auth, and JWT auth.
  • Inline and remote JWKS support: Define both inline and remote JWKS endpoints to automatically fetch and rotate keys from your identity provider on the AgentgatewayPolicy, including TLS options when connecting to a remote JWKS source
  • CSRF: Configure CSRF policies using the traffic field in AgentgatewayPolicy.
  • ExtAuth: ExtAuth with HTTP support and configurable timeout.

Observability

Dynamically configure tracing for agentgateway using the AgentgatewayPolicy frontend field. See the tracing setup guide.

mTLS

TLS encryption for the control plane: Enable server-side TLS encryption for the control plane.

mTLS for agentgateway proxy connections: Support for mTLS listeners is now available, which includes configuring:

  • Cipher suites
  • Minimum and maximum TLS versions
  • Certificate validation

To configure mTLS, use the spec.frontend.tls fields in AgentgatewayPolicy.

Ingress to Gateway API migration

If you are currently running Ingress Nginx to support the Kubernetes Ingress API, the ingress2gateway tool can help you migrate to Gateway API by translating your existing Ingress manifests into Gateway, HTTPRoute, and implementation-specific policy resources. The tool can emit resources tailored for agentgateway. See the ingress to agentgateway migration guide.

🗑️ Deprecated or removed features

No support for kgateway APIs

As previously mentioned, the kgateway APIs are no longer supported for agentgateway version 2.2 and later.

Inference Extension in 2.2.1

Version 2.2.0 includes an inference plugin regression due to GitHub issue #13456. Users of this plugin should not upgrade to v2.2.0 and should instead wait for the upcoming v2.2.1 patch release.

🔮 Future releases

Note that version 2.2 of agentgateway on Kubernetes is the last version to use the kgateway control plane. The next release plans to standardize the versioning for standalone agentgateway and agentgateway on Kubernetes.

As such, the documentation in this agentgateway.dev website is for version 2.2 and later. Version 2.1 documentation is no longer available, as it was previously on the kgateway.dev website.

Agentgateway assistant

Ask me anything about agentgateway configuration, features, or usage.

Note: AI-generated content might contain errors; please verify and test all returned information.

↑↓ navigate select esc dismiss

What could be improved?