Changelog¶

AI Gateway Helm chart v0.1.1¶

March 31, 2025

🆕 Features

• Adds readiness configuration and endpoint for AI Gateway Core image.

• Updates AI Gateway Core image to v1.0.2.

• Updates AI Gateway Processors image to v1.0.1.

✅ Compatibility:

• Kubernetes version 1.25.0-0 or higher

📦 Helm registry:

• oci://private-registry.f5.com/aigw/aigw:0.1.1

AI Gateway Core v1.0.2¶

March 31, 2025

🐛 Bug fixes:

• Fix the core not rejecting request/response for parallel stages: if a stage was parallel, the core would not reject request/response, even if a processor rejected it.

✅ Compatibility:

• Kubernetes version 1.25.0-0 or higher

📦 Container images:

• private-registry.nginx.com/aigw/aigw:v1.0.2

AI Gateway Core v1.0.1¶

March 28, 2025

✨ Highlights:

• The release includes a number of bug fixes.

🐛 Bug fixes:

• Improve error messages in logs: make the messages more descriptive and user-friendly.

• Fix the readiness probe health check: previously, the probe could have succeeded even if the core wasn’t ready to receive the traffic.

• Return 502 Bad Gateway status when the core doesn’t select any service for a request. Previously, the core would return 404.

• Ensure the core adds an error to an OTel trace for a span that corresponds to the profile sequential stage.

• Ensure the core validates the processor parameters (the params field of the processor configuration).

• Ensure the core returns a 500 Internal Server Error when a processor executed in a parallel stage modifies the client request or LLM response. Such modifications are not allowed. Previously, the core would disregard any modifications, but incorrectly proceed further with any processing.

• Ensure the core fails to start if aigw start is invoked without the config file path.

• Do not print usage if the core terminates with an error.

• Ensure the core returns JSON errors when it cannot match a request against any route. Previously, it would return plain text errors.

• When a service returns an error, ensure the core propagates the HTTP status code to the client. Previously, the core would always return the 500 status code.

• Ensure the core returns errors to clients in JSON using the OpenAI-like error format. Previously, for some cases, the core would return errors in different format.

• Ensure when the core is being shutdown, it exports any pending unexpected transaction. Note: if transactions are not exported in 30s (default terminationGracePeriodSeconds pod parameter), Kubernetes will kill the pod, and, as a result, the unexported transactions will get lost.

• Improve validation of the path field in routes.

• Change the minimal acceptable TLS version for connections to processors and services from v1.3 to v1.2.

✅ Compatibility:

• Kubernetes version 1.25.0-0 or higher

📦 Container images:

• private-registry.nginx.com/aigw/aigw:v1.0.1

AI Gateway Processors v1.0.1¶

March 28, 2025

✨ Highlights:

• Minor optimizations for the language-id and prompt-injection processors

• Improved validation of model pipeline parameters

🐛 Bug fixes:

• Fixed an issue with processors server not utilizing worker processors when WEB_CONCURRENCY is set

✅ Compatibility:

• Kubernetes version 1.25.0-0 or higher

📦 Container images:

• private-registry.nginx.com/aigw/aigw-processors-f5:v1.0.1

AI Gateway v1.0.0¶

February 25, 2025

AI Gateway Core v0.2.0¶

January 23, 2025

v0.1.0¶

November 13, 2024

✨ Highlights:

F5 AI Gateway routes generative AI traffic to an appropriate Large Language Model (LLM) backend and protects the traffic against common threats, which includes:

• Inspecting and filtering client requests and LLM responses

• Preventing malicious inputs from reaching an LLM backend

• Ensuring that LLM responses are safe to send to clients

• Protecting against leaking sensitive information

AI Gateway is a standalone product that runs on Kubernetes.