2.24.4

Bug Fixes and Minor Changes

• Convert the HTTP status_code to integer before emitting to statsd (#1311).
• Allow configuration of the Marqo host name in run_marqo.sh (#1314). You can specify the host name via the environment variable MARQO_HOST now.

2.24.3

Bug Fixes and Minor Changes

• Revert the pagination fix for search using the disjunction retrieval method(#1310).

2.24.2

Bug Fixes and Minor Changes

• Fix pagination for search using the disjunction retrieval method(#1303).
• Escape \ and " characters in typeahead queries(#1306).

2.24.1

Bug Fixes and Minor Changes

• Pin the log level of opentelemetry metrics to INFO(#1304).
• Fix a timeout issue for search requests(#1305).

2.24.0

New features

• Support typeahead search(#1289). You can now get intelligent query suggestions based on partial user input to help users find relevant content faster and reduce zero-result searches. For details, please refer to this API document and this recipe. This is supported only for indexes created with Marqo 2.23.0 or later.
• Support caching embeddings of base64-encoded images(#1285). This feature greatly improves performance by cutting the vectorisation time of commonly reused images from ~10 ms down to ~1 ms.

Bug fixes and minor changes

• Fix an issue where facet counts were incorrect when using the lexical retrieval method with filters and multiple OR phrases (#1282).
• Fix an issue where collasing search stops working when the collapse field is not included in attributesToRetrieve search parameter (#1294).
• Fix metric emission for the get-batch and delete-batch endpoints (#1300).
• Ensure access log level now inherits from the root log level (#1301).
• Reduce verbosity of some logs emitted from HybridSearcher (#1299).

2.23.2

Bug Fixes and Minor Changes

• Fix totalHits count for collapsed searches(#1292). When a collapse field is provided in search parameters, the totalHits now reflects the number of unique values for that field (i.e., the number of groups), rather than the raw number of matching documents.

2.23.1

Bug Fixes and Minor Changes

• Add "allowMissingDocuments" and "allowMissingEmbeddings" parameters to the recommend and search with context endpoints (#1287). Users can set these parameters to true to skip errors when documents or their embeddings are missing.
• Relax the restrictions on the "queryTensor" field in hybrid search when using retrievalMethod="disjunction" and rankingMethod="rrf" (#1287). This field can now accept None or an empty dictionary, making hybrid search more flexible.

2.22.2

Bug Fixes and Minor Changes

• Add "allowMissingDocuments" and "allowMissingEmbeddings" parameters to the recommend and search with context endpoints (#1287). Users can set these parameters to true to skip errors when documents or their embeddings are missing.
• Relax the restrictions on the "queryTensor" field in hybrid search when using retrievalMethod="disjunction" and rankingMethod="rrf" (#1287). This field can now accept None or an empty dictionary, making hybrid search more flexible.

2.23.0

New Features

• Configurable stemming for text fields (#1273). You can now control stemming on a per-field basis in a unstructured index. Stemming influences full-text search behavior. Filters are unaffected. Please refer to this document for detailed guidance.

• Collapse fields (variant grouping) (#1276, #1277). Collapse field groups search results by a field (e.g., product_id) and returns only one "top" document per group. It’s perfect for product variants (sizes, colors, SKUs) because you can show the "top" variant per product. This feature is available for hybrid search in unstructured indexes. Please refer to this cookbook for more details.

2.22.1

Bug Fixes and Minor Changes

• Remove the concurrency parameter from context documents in search API (#1268)
• Emit generic high cardinality metrics (#1272)
• Exclude document-operation-executor config from Vespa bootstrapping (#1271)