- Fix bug in
FunctionInvocationnot using per-invocation type conversion when callingwithResultType. - Fix bug in Global Hooks not being invoked under certain circumstances.
- Add fluent returns to
ChatHistoryaddXMessagemethods. - Add user agent opt-out for OpenAI requests by setting the property
semantic-kernel.useragent-disabletotrue. - Add several convenience
invokePromptAsyncmethods toKernel. - Allow Handlebars templates to call Javabean getters to extract data from invocation arguments.
- Improve thread safety of
ChatHistory.
- Add JDBC vector store
- Add custom type Conversion example,
CustomTypes_Example - Dependency updates and pom cleanup
- Documentation updates
- Add ability to use image_url as content for a OpenAi chat completion
- As part of this
ChatMessageTextContentandChatMessageImageContentwas added that extends the existingChatMessageContentclass.ChatMessageContentfor now defaults to a text content type for backwards compatibility. However, users are encouraged to migrate to using the builders onChatMessageTextContentto create text based chat messages. - Constructors of
ChatMessageContentwere also modified to support this change.
- As part of this
- Added preliminary hugging face implementation that is still in development/beta.
- Added Gemini support
- Added OpenTelemetry spans for OpenAI requests
- Update the user agent for OpenAI requests
- Move XML parsing classes to implementation package as they are not expected to be used by users.
- Reorganized the repository when moving to the new Github location
- Removed non-Java files
- Update readmes
- Update build scripts
- Bring back Spring example project
- Fix bug with removing new lines on function parameters on Windows
- Fix bug forming serializing arguments to tool calls
- Fix bug appending plugin name to tool calls
- Improve exception handling in OpenAIChatCompletion
- Upgrade azure-identity to 1.12.1
- Remove fixed netty version in bom
- Upgrade azure-ai-openai to 1.0.0-beta.8
ChatHistoryno longer has a default message, see below for more details.
- Allow setting deployment name in addition to modelId on AI services.
- Remove default message of "Assistant is a large language model" from ChatHistory
- This is a breaking change if you were relying on the default message in your code
- Add InvocationReturnMode and rework OpenAi chat completion to allow configuring what data is returned from Chat requests
- Reorganize example projects and documentation structure.
- Number of sample updates and bug fixes.