Added #

• Integration regression coverage for stream boundary resilience in tool-call loops (read_file -> write_file -> final assistant response)

Changed #

• OllamaStreamConverter.toLLMStream() now uses boundary-safe NDJSON line framing across transport chunks
• Stream parsing now uses bounded malformed-line retries (3 consecutive malformed non-empty lines) and throws explicit LLMApiException after budget exhaustion instead of silently dropping indefinitely
• maxToolAttempts default increased from 25 to 90
• Bumped llm_core dependency to ^0.1.9

Added #

• OllamaMessageConverter.messagesToOllamaJson() for list-aware conversion; derives tool_name from toolCallId via preceding assistant's tool_calls or synthetic ID parsing
• Fallback chain for tool_name: lookup by id, parse synthetic tool_N_name, or send tool_call_id only (Ollama supports both)
• Thorough tool response integration tests (15 cases) validating stream contract, deterministic results, error handling, tool chains, and Ollama-specific behavior

Changed #

• Tool messages now converted via messagesToOllamaJson(); tool_name encapsulated in Ollama layer, derived from toolCallId
• Replaced per-message toJson() with list-aware messagesToOllamaJson() for correct tool message conversion

Added #

• batchEmbed() implementation: delegates to existing batch-capable embed() (Ollama /api/embed accepts an array of inputs).

Fixed #

• Ensured Ollama tool calls always produce LLMToolCall instances with non-null, non-empty id values, synthesizing IDs when Ollama does not provide them.
• Aligned tool-calling behavior with llm_core's toolCallId validation so that tool execution no longer fails with Tool message must have toolCallId when used together with llm_core.

Added #

• Builder pattern for OllamaChatRepository via OllamaChatRepositoryBuilder for complex configurations
• Support for StreamChatOptions in streamChat() method
• Support for chatResponse() method for non-streaming complete responses
• Support for RetryConfig and TimeoutConfig for advanced request configuration
• Input validation for model names and messages

Changed #

• streamChat() now accepts optional StreamChatOptions parameter
• Improved error handling and retry logic
• Enhanced documentation

Added #

• Initial release
• Ollama backend implementation for LLM interactions:
  • Streaming chat responses
  • Tool/function calling support
  • Vision (image) support
  • Embeddings
  • Thinking mode support
  • Model management (list, pull, show, version)
• Full compatibility with Ollama API