- introduce copy() method to the ChatOptions.
- make sure that the DefaultChatClientRequestSpec takes a copy of the input chat options to prevent multation.
- add a OpenAiChatClientMultipleFunctionCallsIT to reproduce the problem and verify the solution.
Resolves#1064
* Remove inheritance from HashMap
* No more subclasses per model provider
* Builder class for ChatResponse
* Fix the AbstractResponseMetadata#AI_METADATA_STRING parameter order
* ChatResponseMetadata ignore Null values.
- re-assign the model on withSystemInstruction return.
- deprecate the gemini-pro-vision model and replace the default model to gemin-pro-1.5
- update docs.
- improve system message handling implementation.
Resolves#1030
Co-authored-by: Christian Tzolov <ctzolov@vmware.com>
- Moves these from free-text key/value pairs to interface
- Enables programmatic use for evaluation and observability
Signed-off-by: Thomas Vitale <ThomasVitale@users.noreply.github.com>
- Add ToolCall to AssistantMessage.
- Rename FunctionMessage to ToolResponseMessage and add id and name fields.
- Refactor OpenAiChatModel's function calling handling.
- Prompt copy now copies the AssistantMessage and ToolResponseMessage contents.
Other ChatModel implementations to adopt these changes in subsequent commits
- add new spring-ai-vertex-ai-embedding project.
- add VertexAiTextEmbeddingModel and VertexAiMultimodalEmbeddingMode with related options configuration classes.
- add ITs
- add auto-configuraiton and boot starters.
- register to BOM.
- add documentation.
- add multimodal embedding documentation
- extend the Embedding metdata so that it can keep references to the source document's data, Id, mediatype
Resolves#1013
Related to #1009
- add dedicated groq chat page in the documentation.
Explain how to re-configure the OpenAI client for accessing the Groq chat completion endpoint.
- Doc: order the Chat and Embedding items in alphabetical order
- Add Groq ITs.
Resolves#996
- add StreamEven API domain model for reliably parsing stream events.
- add StreamHelper#mergeToolUseEvents to aggregate partial tool use jsons into a list of ContentBlocks.
- add StreamHelper#eventToChatCompletionResponse to convert Flux<StreamEvents> into Flux<ChatCompletionResponse>.
- Rename MediaContent -> ContentBlock, RequestMessage -> AnthropicMessage, ChatCompletion -> ChatCompletionResponse.
- Improve tests and docs.
* Extend ChatResponseMetadata for Anthropic (blocking, streaming)
* Add ChatResponseMetadata for Mistral AI (blocking)
* Extend ChatResponseMetadata for OpenAI (blocking)
* Deprecate gpt-4-vision-preview and replace its usage in tests because OpenAI rejects the calls (see: https://platform.openai.com/docs/deprecations)
Fixes gh-936
Signed-off-by: Thomas Vitale <ThomasVitale@users.noreply.github.com>
- OpenAiApi: add StreamingOptions class and ChatCompletionRequest#streamingOptions field.
- add OpenAiChatOption#withStreamingUsage(boolean) to set/unset the StreamingOptions.
- add a boolean (get/set)StreamUsage() to OpenAiChatOptions that internally set the SstreamOptions.
Later allows the "spring.ai.openai.chat.options.stream-usage" property.
- update the OpenAI property documentation.
Co-authored-by: Christian Tzolov <ctzolov@vmware.com>
- upadate bedrockruntime version to 2.26.7 and align it with aws sdk dependecy.
- minor ITs configuration, consistency and statbility fixes
Use the EnvironmentVariableCredentialsProvider instead of the Profile credential provider.
- update the Bedrock getting started documentation.
- update the Genemi Prerequisites docs.
- Added additional support for ZhiPuAi models including vision model GLM-4V
- Default model changed to GLM_4_Air and update documentation
- Added additional unit and integration tests
- Add documentation for vision model
* Align model naming with Mistral AI documentation
* Add missing model for mixtral 22B
* Deprecate mistral-medium-latest because Mistral will remove it soon
Signed-off-by: Thomas Vitale <ThomasVitale@users.noreply.github.com>
Some LLM providers, such as Groq and OpenRouter, are marketed as OpenAI API compatible.
However, they often lack full support for the API specification.
This PR allows to use the simpler message format if no media artifacts are assigned.
