CHANGELOG

Changelog for package bob_llm

1.0.3 (2026-04-30)

  • Refactor monolithic prompt processing into event-driven architecture (removed polling timer).

  • Implement O(n) chat history trimming for improved performance during long conversations.

  • Add robust stream recovery with automatic retry for failed API connections.

  • Implement token buffering for smoother streaming and reduced ROS message overhead.

  • Integrate ThreadPoolExecutor for isolated tool execution with configurable timeouts.

  • Add tool_timeout parameter (default 60s) for robust skill execution.

  • Improve multimodal content preservation (images) and smart text extraction for logging.

  • Refine tool budgeting to count only successful calls against the limit.

  • Add clean node shutdown logic with executor and thread cleanup.

  • Add comprehensive integration test suite for LLM flow verification.

  • Fix chat history reasoning issues and improve turns management.

  • Implement definitive zero-latency SSE streaming via iter_lines.

  • Fix UTF-8 encoding for special characters in raw byte streams.

  • Optimize chat UI refresh rate for better human perception.

  • Integrate tool call detection in reasoning stream for faster response.

  • Restore 100% flake8/PEP8 compliance and single quote enforcement.

  • Refactor main interaction loop for robust synchronous execution.

  • Fix JSON prompt handling and enhance system prompt file support.

  • Add support for dynamic system_prompt_file parameter loading.

  • Implement dynamic parameter reconfiguration for LLM client.

  • Add optional eof parameter to signal end of stream on llm_stream.

  • Add tool_choice parameter for dynamic tool call control.

  • Enhance tool execution logging with result previews.

  • Remove prefix v1 from chat API path for standard compatibility.

  • Add support for Agentskills specification and modular tools.

  • Add native Qdrant vector database tools with env configuration.

  • Refactor Agent Skills to follow progressive disclosure patterns.

  • Fix Race Condition in LLMNode node-to-client initialization.

  • Implement soft limit for tool calls with final response hint.

  • Add llm_reasoning topic for live thinking content support.

  • Update OpenAI client to extract reasoning_content from chunks.

  • Enforce mandatory discovery in tool docstrings for safety.

  • Improve type safety in backend_clients with proper annotations.

  • Add premium interactive terminal chat client with boxed UI.

  • Clean up legacy scripts and modernize README documentation.

  • Contributors: Bob Ros

1.0.2 (2026-02-01)

  • Full ROS 2 Rolling and Humble compliance (fixed linter issues)

  • Standardized import ordering and quote usage

  • Contributors: Bob Ros

1.0.1 (2026-01-26)

  • Fix 270+ linter and style issues for ROS2 compliance

  • Fix package.xml schema validation

  • Standardize docstrings and copyright headers

  • Contributors: Bob Ros

1.0.0 (2025-11-25)

  • Initial release of bob_llm

  • Contributors: Bob Ros