The dependency on spring-webflux was unnecessary, there was no code
inside spring-ai-retry that made use of it.
Also remove 'optional' from slf4j-api dependency (it's not optional,
if you don't have it on classpath, RetryUtils will fail to load).
Add explicit dependency on spring-webflux for modules that directly
import WebClient.
Fixes#3307
Signed-off-by: Piotr Kubowicz <piotr.kubowicz@gmail.com>
(cherry picked from commit cecc046c26)
Unit test to veriy OpenAiChatOptions.fromOptions.webSearchOptions
Fixes: 3377
Signed-off-by: lambochen <lambochen@yeah.net>
(cherry picked from commit 4621c812cc)
- Fixed an issue where the `from` method of `org.springframework.ai.ollama.OllamaChatModel` could not work correctly when a tool call occurred while using Ollama.
Signed-off-by: Sun Yuhan <1085481446@qq.com>
* feat(anthropic): add support for claude-opus-4-0 and claude-sonnet-4-0 models
- Introduce `CLAUDE_OPUS_4` and `CLAUDE_SONNET_4` to `ChatModel` enum
- Update documentation to include new model options
Signed-off-by: Alexandros Pappas <apappascs@gmail.com>
* add model aliases url to anthropic documentation
Signed-off-by: Alexandros Pappas <apappascs@gmail.com>
---------
Signed-off-by: Alexandros Pappas <apappascs@gmail.com>
- Convert non-primitive metadata values to JSON strings in ChromaApi.AddEmbeddingsRequest for compatibility
- Add tests to verify metadata conversion and complex metadata handling in Chroma vector store integration
- Ensure OpenAiChatModel always returns annotations as a list of maps in metadata
- Add test dependency on spring-ai-advisors-vector-store for advisor-related tests
- Add json-unit-assertj for improved JSON assertions and validating JSON content in tests
Signed-off-by: Christian Tzolov <christian.tzolov@broadcom.com>
Replace spring-ai-client-chat dependency with spring-ai-model in model implementations
and memory repositories, and with spring-ai-commons in document readers. This change
improves the dependency structure by having components depend on the appropriate
abstraction level.
Additional changes:
- Add slf4j-api dependency to pdf-reader and spring-ai-retry
- Move spring-ai-client-chat to test scope in spring-ai-ollama
- Fix XML formatting in some pom.xml files
Signed-off-by: Christian Tzolov <christian.tzolov@broadcom.com>
samples() duplicates N(), both setting the same field.
Keeping only N() simplifies the builder API and aligns with ImageOptions.
Signed-off-by: Jhzlo <kjh0010703@naver.com>
Replaces redundant or conflicting stylePreset assignments with a single mergeOption call using runtime style and default stylePreset.
Signed-off-by: Jhzlo <kjh0010703@naver.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 fix overrides the adviseStream method to use ChatClientMessageAggregator to properly aggregate streaming chunks before storing them in memory, similar to how PromptChatMemoryAdvisor handles streaming responses
- Added test
Signed-off-by: Mark Pollack <mark.pollack@broadcom.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: #2168
- Change property name from 'taskType' to 'task_type' in VertexAiEmbeddingUtils to match Google API expectations
- Add integration tests to verify task type behavior matches Google SDK
- Add missing auto truncate option copying in VertexAiTextEmbeddingOptions
Signed-off-by: Soby Chacko <soby.chacko@broadcom.com>
- Enhance MethodToolCallback to properly handle generic types by using parameterized types
- Add unit tests for generic type handling (List, Map<String,Integer>, nested generics)
- Add integration tests for both Anthropic and OpenAI clients to verify tool calls with generic argument types
Resolves#2462
Signed-off-by: Christian Tzolov <christian.tzolov@broadcom.com>
Add integration tests for BedrockNovaChatClient that verify:
- Tool annotation for method without input arguments.
- Supplier-based tool calling implementation.
Related to #1878
Signed-off-by: Christian Tzolov <christian.tzolov@broadcom.com>
- For all the model response API objects, add `@JsonIgnoreProperties(ignoreUnknown = true)`
which provides flexibility to ignore any unknown properties which come as part of the response
Resolves#3026
Signed-off-by: Ilayaperumal Gopinathan <ilayaperumal.gopinathan@broadcom.com>
- Introduced mutate() methods to OpenAiApi and OpenAiChatModel, enabling creation of new builder instances from existing objects.
- Allows safe modification and copying of configuration for APIs and models.
- Refactored internal fields and getters to support mutation/copy patterns.
- Updated integration tests to leverage mutate for dynamic client creation.
This PR adds support for retrieving web search annotations from the OpenAI API, as described in their [web search documentation](https://platform.openai.com/docs/guides/web-search). This allows us to access citation URLs and their context within generated responses when using models like `gpt-4o-search-preview`.
**Changes:**
* Added `annotations` (with `Annotation` and `UrlCitation` records) to `ChatCompletionMessage` in `OpenAiApi.java`.
* Updated `OpenAiChatModel` to populate the `annotations` field (via metadata) for both regular and streaming responses.
* Added integration tests (`webSearchAnnotationsTest`, `streamWebSearchAnnotationsTest`) to `OpenAiChatModelIT.java`.
* Added `GPT_4_O_SEARCH_PREVIEW` and `GPT_4_O_MINI_SEARCH_PREVIEW` to `OpenAiApi.ChatModel`.
* Added `WebSearchOptions` and related records to `OpenAiApi`.
* Minor updates to `ChatCompletionRequest` and its `Builder`.
Resolves#2449
Signed-off-by: Alexandros Pappas <apappascs@gmail.com>
* Adds reasoningEffort field to AzureOpenAiChatOptions builder, copy, equals, hashCode
* Propagates value to ChatCompletionsOptions
Fixes#2703
Signed-off-by: Andres da Silva Santos <40636137+andresssantos@users.noreply.github.com>
Remove circular dependencies:
Move utility classes from various packages to dedicated support packages:
- Move ToolCallbacks from ai.tool to ai.support
- Move UsageUtils to UsageCalculator in ai.support
- Move tool.util to tool.support
- Create new ToolDefinitions utility class
Support packages for the classes in question is more idomatic in spring than util packages.
Signed-off-by: Mark Pollack <mark.pollack@broadcom.com>
Without this fix during the stream event handling when `EventType.MESSAGE_STOP` occurs, the latest content block was resent again and it caused to the additional tool call(if it was the latest event)
[anthropic] Replace `switchMap` -> `flatMap` to avoid cancellation of the original request
Previously, internalStream used switchMap to process ChatCompletionResponses,
which caused the active stream (including potential recursive calls) to be
canceled whenever a new response arrived. This led to incomplete processing
of streaming tool calls and unexpected behavior when handling tool_use events.
Replaced switchMap with flatMap to ensure that each response is fully processed
without being interrupted, allowing recursive internalStream calls to complete
as expected.
Without this fix during the stream event handling when `EventType.MESSAGE_STOP` occurs, the latest content block was resent again and it caused to the additional tool call(if it was the latest event)
Signed-off-by: Mikhail Mazurkevich <Mikhail.Mazurkevich@jetbrains.com>
