LangChain’s AzureAISearchVectorStore expects specific field names (content, content_vector) that don’t match Azure AI Search’s default field names (content_text, content_embedding). Here’s how to bridge the gap.
The Problem
AzureAISearchVectorStore provides excellent support for hybrid and semantic queries against Azure AI Search. However, when connecting to an existing Azure AI Search index, it doesn’t work out of the box.
The issue is a field name mismatch:
| LangChain Expects | Azure AI Search Creates |
|---|---|
| content | content_text |
| content_vector | content_embedding |
LangChain hardcodes these field names internally, and currently there’s no configuration option to customize them.
The Solution
Add output field mappings to your indexer configuration to map Azure’s field names to what LangChain expects:
Then configure the vector store:
Feature Request
It would be helpful if AzureAISearchVectorStore supported custom field name mappings in the configuration, similar to how other vector stores handle this. This would eliminate the need for indexer-level workarounds.
this is a great request. Would you mind raising an issue in GitHub? We would appreciate any types of contributions here.