Changelog

All notable changes to the Chaukas SDK will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[0.2.0] - TBD

LangChain (langchain>=0.1.0,<2.0) - 17/19 events (~95% coverage)
- Full event capture for LangChain chains, agents, and tools
- Support for LCEL (LangChain Expression Language) pipelines
- Automatic retry detection and tracking
- MCP protocol support (via mcp_server metadata)
- Streaming output capture (via on_text callback)
- Custom events (via on_custom_event callback)
- Examples: basic chains, agents with tools, LCEL pipelines, RAG, streaming
- Production ready
- Note: POLICY_DECISION and STATE_UPDATE require app-level instrumentation
- Note: LangChain 1.x requires Python ≥3.10

OpenAI Agents - Event coverage updated to 18/19 events (94.7%)
- RETRY events are not supported for OpenAI Agents SDK
- Reason: OpenAI SDK performs retries internally within its HTTP client layer, making retry attempts invisible to external instrumentation
- All other event types fully supported including ERROR events (captured after SDK retries are exhausted)
- Removed retry-related examples from OpenAI documentation

OpenAI Agents - Type safety improvements
- Replaced fragile string comparisons with proper isinstance() type checking
- Added imports for OpenAI tool types: FileSearchTool, WebSearchTool, CodeInterpreterTool, HostedMCPTool, LocalShellTool, ComputerTool
- Added imports for OpenAI exception types: APIError, RateLimitError, APIConnectionError, Timeout, AuthenticationError
- Fixed undefined variable bug in on_llm_end hook
- Fixed invalid finish_reason values (removed non-existent 'content_policy' and 'moderation' checks)
OpenAI Agents - Code quality improvements
- Reduced code duplication (~48 lines removed, 24% reduction)
- Created helper methods for attribute access (_get_agent_id, _get_agent_name, _get_model, etc.)
- Consolidated async/sync method pairs with shared helpers
- Improved tool span lookup from O(n) to O(1) performance
- Added _parse_llm_response() for safe response validation
- Moved asyncio import to module level for better performance
OpenAI Agents - Enhanced data access tracking
- Added support for LocalShellTool (datasource: "local_shell")
- Added support for ComputerTool (datasource: "computer")
- Now tracks 5 data access tool types (previously 3)
OpenAI Agents - Error handling improvements
- Fixed ERROR events not being emitted in error scenarios
- Added AGENT_END events (FAILED status) when errors occur
- Improved agent stack management for proper cleanup on errors

Initial Release - Open-source SDK implementing the chaukas-spec for AI agent observability

OpenAI Agents (openai-agents>=0.5.0,<1.0.0) - 19/19 events (100% coverage)
- Full event capture including sessions, agents, LLM calls, tools, MCP protocol, retries, errors, and more
- Examples: basic usage, comprehensive workflows, retry patterns, MCP integration
- Production ready
CrewAI (crewai>=1.4.1,<2.0.0) - 19/19 events (100% coverage)
- Event bus integration for complete observability
- Support for multi-agent handoffs, knowledge sources, guardrails, and flows
- Examples: basic crews, comprehensive scenarios, retry patterns, tools, knowledge, guardrails, and flows
- Production ready
Google ADK (latest) - 5/19 events (26% coverage)
- Basic agent and LLM tracking
- Under active development

19 Event Types - Complete implementation of chaukas-spec
- Agent lifecycle: SESSION_START/END, AGENT_START/END, AGENT_HANDOFF
- Model operations: MODEL_INVOCATION_START/END
- Tool execution: TOOL_CALL_START/END, MCP_CALL_START/END
- I/O tracking: INPUT_RECEIVED, OUTPUT_EMITTED
- Operational: ERROR, RETRY, POLICY_DECISION, DATA_ACCESS, STATE_UPDATE, SYSTEM
Zero-Configuration Instrumentation
- One-line setup: chaukas.enable_chaukas()
- Automatic framework detection and patching
- No code changes required
Distributed Tracing
- Session/trace/span hierarchy
- UUID7 event IDs for time-ordered tracking
- Parent-child relationship tracking
Flexible Output
- File mode: JSONL output for local development
- API mode: gRPC streaming to Chaukas platform
- Configurable batching and flushing
Production Ready
- Comprehensive test suite
- Type-safe with full type hints
- Extensive examples and documentation
- Apache 2.0 licensed