API7 Gateway 3.8.0

AI Proxy — Implemented new identifier fields logging and summaries for collecting LLM request and response content.

Release Date: 2025-04-22

Features

AI Proxy (opens in Plugin Hub docs)
- Implemented new identifier fields logging and summaries for collecting LLM request and response content.
AI Proxy Multi (opens in Plugin Hub docs)
- Supported HTTP POST method for health checks.
- Displayed a record at the warning log level if the fallback was triggered.
- Implemented a new identifier field for collecting LLM request and response content.
AI Aliyun Content Moderation (opens in Plugin Hub docs)
- Added support for streaming (HTTP SSE) scenarios.
AI Prompt Guard (opens in Plugin Hub docs)
- Introduced new plugin.
Kafka Logger (opens in Plugin Hub docs)
- Supported reading the user request and response body content cached in ctx and pushing it to the specified Kafka service.

Supported using values from an array object within the user request body in routing conditions. While post_arg was already permitted in routing conditions, this update introduces support for array scenarios for the type field's location.
Supported the recording of additional AI request context information within access logs.
Supported adding a request type identifier to Prometheus metrics to collect more granular data on AI requests.

Added detailed statistics for service name and route name in status code alert messages.
Optimized plugin configuration: When configuring Traffic Split (opens in Plugin Hub docs) plugin on a service or route, verify that the upstream_id configured indeed belongs to the current service's upstreams.
Aligned the upstream connection configuration field names with the API in published services.
Introduced a form-based UI for CORS (opens in Plugin Hub docs) plugin configuration.

Fixed issue: When the upstream type was Kubernetes service discovery, nodes in health checks were not updated after node changes, resulting in many health check failure logs.
Fixed issue: An error occurred when the shared_size parameter for Kubernetes service discovery was adjusted to 100m.#11857
Fixed issue: APISIX/API7 Enterprise with Kubernetes discovery will fail after token file expires. #11779
Fixed issue: After updating the prefix of Vault secret provider, the data plane continued to use the old configuration.

Fixed issue: The retries field in the upstream was incorrectly set as required.
Fixed issue: An error occurred on the published service page due to lack of service template permissions.
Fixed issue: Updating the service discovery service name in published services did not take effect.
Fixed issue: Creating multiple records with rapid clicks.
Fixed issue: High database CPU utilization when using large scale of consumers.
Optimized the wildcard configuration method for Permission Policy:
- Wildcard configuration using * is now supported in authorization statements, with the same meaning as the current <.*>；
- The asterisk * is restricted and not allowed in any resource ID fields to prevent authorization failures.