What should we do?
Use the global Vertex AI endpoint (aiplatform.googleapis.com) in the IDP extraction connector's VertexAiClient instead of the default regional endpoint ({region}-aiplatform.googleapis.com).
The change is adding .apiEndpoint("aiplatform.googleapis.com") to the VertexAiGeminiChatModel.builder() call in VertexAiClient.java.
File: connectors/idp-extraction/src/main/java/io/camunda/connector/idp/extraction/client/ai/VertexAiClient.java
Why should we do it?
Newer Gemini models (3.x series — e.g. Gemini 3.5 Flash, 3.1 Pro Preview, 3.1 Flash Lite) are only accessible via global endpoint routing. Without this change, users of the IDP extraction connector cannot use these models.
The global endpoint is backward compatible:
- Older models (Gemini 2.5, 2.0, etc.) continue to work unchanged
- The
.location() parameter still governs data residency via the request path
- The global endpoint routes requests internally to the correct region
What should we do?
Use the global Vertex AI endpoint (
aiplatform.googleapis.com) in the IDP extraction connector'sVertexAiClientinstead of the default regional endpoint ({region}-aiplatform.googleapis.com).The change is adding
.apiEndpoint("aiplatform.googleapis.com")to theVertexAiGeminiChatModel.builder()call inVertexAiClient.java.File:
connectors/idp-extraction/src/main/java/io/camunda/connector/idp/extraction/client/ai/VertexAiClient.javaWhy should we do it?
Newer Gemini models (3.x series — e.g. Gemini 3.5 Flash, 3.1 Pro Preview, 3.1 Flash Lite) are only accessible via global endpoint routing. Without this change, users of the IDP extraction connector cannot use these models.
The global endpoint is backward compatible:
.location()parameter still governs data residency via the request path