- Override adviseStream method in VectorStoreChatMemoryAdvisor to properly handle streaming responses
- Add tests to verify the fix works with both normal and problematic streaming scenarios
Fixes#3152
Signed-off-by: Mark Pollack <mark.pollack@broadcom.com>
- Override adviseStream method in VectorStoreChatMemoryAdvisor to properly handle streaming responses
- Add tests to verify the fix works with both normal and problematic streaming scenarios
Fixes#3152
Signed-off-by: Mark Pollack <mark.pollack@broadcom.com>
* The defaultConversationId was configurable, but not used. It’s now being used correctly when a custom defaultConversationId is defined.
* The memory advisors were missing the required configuration of a Schedule due to a default value missing. Now as default Scheduler is used, automatically protecting from blocking. It can be customised via “scheduler()”, replacing the old “protectFromBlocking()” method.
* The new defaultTopK options in VectorStoreChatMemoryAdvisor were documented, but not implemented. That is fixed now.
* The memory advisors were not null-safe. Now they are.
* Improved tests to check the null-safe behaviour.
* Updated the documentation accordingly.
Fixes gh-3133
Signed-off-by: Thomas Vitale <ThomasVitale@users.noreply.github.com>
- Remove deprecated ChatMemory.get(String conversationId, int lastN) method
- Replace AbstractChatMemoryAdvisor with BaseChatMemoryAdvisor interface in api package
- Make constructors private in all memory advisor implementations to enforce builder usage
- Rename CHAT_MEMORY_CONVERSATION_ID_KEY to CONVERSATION_ID and move to ChatMemory interface
- In VectorStoreChatMemoryAdvisor:
- Rename DEFAULT_CHAT_MEMORY_RESPONSE_SIZE (100) to DEFAULT_TOP_K (20)
- Rename builder method chatMemoryRetrieveSize() to topK()
- Remove systemTextAdvise() builder method
- In PromptChatMemoryAdvisor:
- Remove systemTextAdvise() builder method
- Fix bug where only the last user message was stored from prompts with multiple messages
- Enhance logging in memory advisors to aid in debugging
- Add comprehensive tests for all advisor implementations:
- Unit tests for builder behavior
- Integration tests for the various chat memory advisors
Signed-off-by: Mark Pollack <mark.pollack@broadcom.com>
Fixes: #1831
- Document auto-truncation configuration with high token limits
- Add integration tests for auto-truncation behavior
- Include Spring Boot and manual configuration examples
- Test large documents and batching scenarios
Enables proper use of embedding model auto-truncation while avoiding batching strategy exceptions.
Signed-off-by: Soby Chacko <soby.chacko@broadcom.com>
* Update remaining Advisors and related classes to use the new APIs.
* In AbstractChatMemoryAdvisor, the “doNextWithProtectFromBlockingBefore()” protected method has been changed from accepting AdvisedRequest to ChatClientRequest. It’s a breaking change since the alternative was not part of M8.
* MessageAggregator has a new method to aggregate messages from ChatClientRequest. The previous method aggregating messages from AdvisedRequest has been removed. Warning since it wasn’t marked as deprecated in M8.
* In SimpleLoggerAdvisor, the “requestToString” input argument needs to be updated to use ChatClientRequest. It’s a breaking change since the alternative was not part of M8. Same thing about the constructor.
* The “getTemplateRenderer” method has been removed from BaseAdvisorChain. Each Advisor is encouraged to accept a PromptTemplate to achieve self-contained prompt augmentation operations.
* Remove deprecations in ChatClient and Advisors, and update tests accordingly.
* When building a Prompt from the ChatClient input, the SystemMessage passed via systemText() is placed first in the message list. Before, it was put last, resulting in errors with several model providers.
Relates to gh-2655
Signed-off-by: Thomas Vitale <ThomasVitale@users.noreply.github.com>
- Avoid overlapping package names
Changed in spring-ai-commons package from org.sf.ai.model to org.sf.ai.content
Refactor advisor module name to be spring-ai-advisors-vector-store
Moved advisors into org.springframework.ai.chat.client.advisor.vectorstore
- Created top level memory directory
- Create new module spring-ai-model-chat-memory-neo4j and moved neo4j memory classes out of the vectorstore module
Updated neo4j autoconfiguation
- Remove deprecations from models, vector stores and usage
- Deprecations from FunctionCallback and ObservationContext/Convention will be in a separate PR
Models updates
- Remove AbstractToolCallSupport from the models which use ToolCallingManager
- Remove deprecated constructors and their usage
- Remove FunctionCallbackResolver and FunctionCallbacks usage in the models
- Add back deprecations for VectorStoreChatMemoryAdvisor until builder is fixed
- Update OpenAiPaymentTransactionIT to use ToolCallbackResolver in config
Signed-off-by: Ilayaperumal Gopinathan <ilayaperumal.gopinathan@broadcom.com>
Extract common vector store delete tests to base class
This commit extracts shared delete operation tests into a reusable BaseVectorStoreTests class.
This reduces code duplication and provides a consistent test suite for delete operations across
different vector store implementations. The base class includes tests for:
Deleting by ID
Deleting by filter expressions
Deleting by string filter expressions
Most of the vector store implementation now extends this base class and inherits these
common tests while maintaining the ability to add vector store specific tests.
Adding javadoc
Signed-off-by: Soby Chacko <soby.chacko@broadcom.com>
- Change VectorStore.delete() and related implementations to return void instead of Optional<Boolean>
- Remove unnecessary boolean return values and success status checks across all vector store implementations
- Clean up tests by removing redundant assertions
- Implementations continue using runtime exceptions for error signaling
Signed-off-by: Soby Chacko <soby.chacko@broadcom.com>
- Add handling for UUID, TEXT, INTEGER, SERIAL, BIGSERIAL formats in `convertIdToPgType` function.
- Implemented type conversion logic based on the IdType value (UUID, TEXT, INTEGER, SERIAL, BIGSERIAL).
- Add unit tests to validate correct conversion for UUID and non-UUID IdType formats.
- `testToPgTypeWithUuidIdType`: Validates UUID handling.
- `testToPgTypeWithNonUuidIdType`: Validates handling for non-UUID IdTypes.
Signed-off-by: jitokim <pigberger70@gmail.com>
Adds getNativeClient API to VectorStore interface allowing access to the underlying native client implementation.
This change:
- Adds getNativeClient() default method to VectorStore interface returning Optional<T>
- Implements getNativeClient() in all vector store implementations exposing their respective native clients
- Adds integration tests verifying native client access for all implementations
Fixes: #2137
Signed-off-by: Soby Chacko <soby.chacko@broadcom.com>
Add string-based filter deletion alongside the Filter.Expression-based deletion
for vector stores. This provides a more convenient API for simple filter cases
while maintaining the full flexibility of expression-based filtering.
Key changes:
- Add delete(Filter.Expression) and delete(String) methods to VectorStore
- Add default string filter implementation in VectorStore interface
- Implement filter deletion in Chroma/Elasticsearch/PgVector stores
- Add integration tests for both filter APIs across implementations
This extends vector store deletion capabilities while maintaining consistent
behavior across implementations.
Moving BatchingStrategy configuration from individual vector store implementations
to the base AbstractVectorStoreBuilder to reduce code duplication and provide consistent
batching behavior across all vector stores.
The default TokenCountBatchingStrategy is now set in the base builder class.
* Remove use of Document.getContext method from spring-ai-core, use getText
* Remove deprecated ChatOptionsBuilder class
* Remove deprecated FunctionCallingOptionsBuilder class
- Replace Document.getContent() with getText() across all vector store implementations
- Fix incorrect package declarations in package-info.java files
- Make builder constructors private and implement proper builder patterns
- Add @Nullable annotations for better null safety
- Simplify PineconeVectorStore builder API by requiring essential parameters in factory method
- Make static Map fields final
- Clean up code and improve JavaDoc documentation
The changes focus on making the vector store APIs more consistent,
type-safe, and maintainable while following best practices for
builder patterns and null safety.
This commit refactors the builder pattern implementation across all VectorStore
implementations to make the EmbeddingModel a required constructor parameter
rather than an optional builder method. Key changes include:
- Move embeddingModel from being a builder method to a required constructor parameter
- Make embeddingModel final in AbstractVectorStoreBuilder
- Remove redundant validate() methods since EmbeddingModel validation now happens
in constructor
- Update all VectorStore builder instantiations to pass EmbeddingModel in builder
creation
- Add @Nullable annotations to appropriate methods in VectorStore interface
This change improves the API design by:
1. Enforcing that EmbeddingModel is provided at builder creation time
2. Removing the possibility of forgotten EmbeddingModel configuration
3. Simplifying the builder implementation by moving validation to construction
4. Making the dependency on EmbeddingModel more explicit in the API
Breaking Changes:
- VectorStore builders must now be created with an EmbeddingModel parameter
- The embeddingModel() builder method has been removed from all implementations
This commit deprecates builder methods with the "with" prefix in the VectorStoreObservationContext class
and introduces new methods without the prefix for a cleaner API. For example:
withCollectionName() → collectionName()
withDimensions() → dimensions()
withNamespace() → namespace()
The old methods are marked as deprecated for maintaining backward compatibility.
All vector store implementations have been updated to use the new method names.
- Move PgVectorStore and related classes to org.springframework.ai.pg.vectorstore package
- Update builder pattern to use more idiomatic method names (e.g. withSchemaName -> schemaName)
- Deprecate existing constructors and old Builder class in favor of new static builder() method
- Update tests to reflect the new builder style usage
- Update docs
The Document class previously allowed multiple media entries while also having a
text field, leading to ambiguity in content handling. This change enforces a
clear separation between text and media documents to prevent content type
confusion and simplify document processing.
A Document now must contain either text content or a single media entry, but
never both. This aligns with the class's primary use in ETL pipelines where
clear content type boundaries are essential for proper embedding generation and
vector database storage.
Additional architectural changes:
- Document now implements a cleaner API by removing deprecated methods
- Removed MediaContent interface implementation from Document class
- Document.getMedia() now returns a single Media object instead of Collection
- Removed EMPTY_TEXT constant in favor of proper null handling
- Constructor signatures simplified and streamlined
- Builder pattern improved to enforce single content type constraint
The breaking changes include:
- Media is now a single entry instead of a collection
- Content field renamed to text for clarity
- Removed support for mixed content types
- Simplified builder API to prevent ambiguous construction
Prefer using text-related methods over deprecated content methods to
better reflect the actual content type being handled and improve API clarity.
- Since the Document object's reference to the `embedding` is deprecated and will be removed, the VectorStore implementations require a way to store the embedding of the corresponding Document objects
- One way to fix this is, to have the EmbeddingModel#embed to return the embeddings in the same order as that of the Documents passed to it.
- Since both the Document and embedding collections use the List object, their iteration operation will make sure to keep them in line with the same order.
- A fix is required to preserve the order when batching strategy is applied.
- Updated the Javadoc for BatchingStrategy
- Fixed the Document List order in TokenCountBatchingStrategy
- Refactored the vector store implementations to update this change
Resolves #GH-1826
Document
* Introduced “score” attribute in Document API. It stores the similarity score.
* Consolidate “distance” metadata for Documents. It stores the distance measurement.
* Adopted prefix-less naming convention in Document.Builder and deprecated old methods.
* Deprecated the many overloaded Document constructors in favour of Document.Builder.
Vector Stores
* Every vector store implementation now configures a “score” attribute with the similarity score of the Document embedding. It also includes the “distance” metadata with the distance measurement.
* Fixed error in Elasticsearch where distance and similarity were mixed up.
* Added missing integration tests for SimpleVectorStore.
* The Azure Vector Store and HanaDB Vector Store do not include those measurements because the product documentation do not include information about how the similarity score is returned, and without access to the cloud products I could not verify that via debugging.
* Improved tests to actually assert the result of the similarity search based on the returned score.
Signed-off-by: Thomas Vitale <ThomasVitale@users.noreply.github.com>
The system text advise prompt is only joined with the value of the system text if it's non-null and non-empty.
Signed-off-by: Thomas Vitale <ThomasVitale@users.noreply.github.com>
Adds @EnabledIfEnvironmentVariable annotation to integration tests that
use OpenAI embeddings. Tests will be skipped if OPENAI_API_KEY is not set,
making the build process more reliable for contributors who don't have
access to OpenAI services.
ObjectMapper instantiation is costly, so unless its usage
is one-shot, it is better to create a reusable instance
for upcoming usage.
Also, before this commit, serialization of most Kotlin
classes was not supported due to the lack of proper
Jackson KotlinModule detection.
This commit:
- Avoids per invocation ObjectMapper instantiation when
relevant
- Automatically detects and enables well-known Jackson
modules including the Kotlin one
- Removes org.springframework.ai.vectorstore.JsonUtils
which looks not needed anymore
More optimizations are possible like reusing more
ObjectMapper instances, but this could introduce more breaking
changes so this commit intends to be a good first step.
Kotlin tests will be provided in a follow-up commit.
Additional changes:
- Update ModelOptionsUtils to use JacksonUtils.instantiateAvailableModules()
- Add missing license headers
- Add missing author Javadoc comments
* Unify image definition for vector stores in vector-store modules
* Unify image definition for vector stores in spring-ai-testcontainers module
Signed-off-by: Thomas Vitale <ThomasVitale@users.noreply.github.com>
Vector store observations support several key-value pairs, coming from the Spring AI abstractions. Currently, whenever a value is not available (either because not configured by the user or not supported by the vector store provider), span/metrics attributes are generated anyway with value none.
That causes several issues, including an unneeded increase in time series, challenges in alerting/monitoring (especially for integer/double attributes that suddenly are populated with a string), and non-compliance with the OpenTelemetry Semantic Conventions (according to which, attributes should be excluded altogether if there's no value).
This pull request changes the conventions for vector store observations to exclude the generation of span/metrics attributes for optional values which don't have any value.
Signed-off-by: Thomas Vitale <ThomasVitale@users.noreply.github.com>
