Release v0.3.0

Changelog

All notable changes to this project will be documented in this file.

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

[0.3.0] - 2025-12-24

Added

Agents

• ToolAgent: ReAct-style agent with automatic tool tracing
• PlannerAgent: Generates structured execution plans before acting
• AgentResult: Unified result model with events, token usage, and cost tracking
• Plan/PlanStep: Structured plan models with dependency support for parallel execution

Orchestration

• Orchestrator: Executes plans by routing steps to registered agents
• AgentRegistry: Named agent container for multi-agent workflows
• Parallel execution: Steps without dependencies run concurrently via asyncio.gather
• Saga pattern: Automatic rollback with compensating actions on failure

Memory

• MemoryHub: Unified interface to short-term and semantic memory
• MemoryContext: Session-scoped context manager for memory operations
• Short-term backends: InMemory, SQLite, Redis
• Semantic backend: Qdrant with FastEmbed embeddings (memory, local, network modes)

CLI

• agenthelm run: Execute tasks with automatic trace storage
• agenthelm plan: Generate execution plans with --output to save as YAML
• agenthelm execute: Run plans from YAML files with --dry-run preview
• agenthelm chat: Interactive REPL mode
• agenthelm traces list/show/filter/export: View and manage execution traces
• agenthelm mcp list-tools/run: Connect to MCP servers
• agenthelm config: Manage configuration
• --verbose flag for debug logging
• --trace-storage option for custom trace storage path

Observability

• ExecutionTracer: Automatic tracing of tool executions with storage
• OpenTelemetry integration: init_tracing() with Jaeger support
• Cost tracking: Built-in LLM pricing for 20+ models (OpenAI, Anthropic, Mistral, Google, etc.)
• Trace export: JSON, CSV, and Markdown formats

MCP Integration

• MCPClient: Low-level MCP protocol client
• MCPToolAdapter: Wraps MCP server tools as AgentHelm-compatible callables
• Compensation definitions for Saga pattern rollback

Documentation

• New Quick Start guide
• Comprehensive CLI reference
• Updated API reference with all new classes
• Working examples in examples/ directory

Changed

• Migrated from custom agent implementation to DSPy-native agents
• Trace storage now defaults to ~/.agenthelm/traces.db
• CLI now uses Click for argument parsing
• Documentation now uses mike for versioned docs

Fixed

• Proper async context manager cleanup for MCP connections
• PYTHONUNBUFFERED environment for Windows MCP compatibility

[0.2.0] - 2025-11-03

Added

• Storage Abstraction Layer: Introduced a flexible storage system with a BaseStorage abstract base class.
• JSON and SQLite Storage: Implemented JsonStorage (default) and SqliteStorage for persisting agent execution
  traces.
• CLI Trace Explorer: Added a new agenthelm traces command group to the CLI.
• traces list command: List recent traces with pagination and JSON output option.
• traces show command: Show detailed information for a specific trace.
• traces filter command: Filter traces by various criteria like tool name, status, date, execution time, and
  confidence score.
• traces export command: Export filtered traces to JSON, CSV, or Markdown formats.
• Observability Documentation: Added docs/observability.md with detailed information on the new features.
• Observability Example: Added a new example in examples/observability_example/ to demonstrate the new storage and
  CLI features.

Changed

• Standardized Logging: Replaced all print() statements with Python's standard logging module for better control
  over output verbosity.
• CLI Entry Point: The main CLI entry point is now orchestrator.cli:app.
• Dependencies: Added tabulate as a new dependency for table formatting in the CLI.
• Project Structure: Moved storage-related classes to the orchestrator/core/storage/ directory.

Fixed

• Corrected various import errors and fixed linter warnings identified by ruff.
• Fixed an issue where the SQLite database file was being deleted after table creation in the example agent.
• Resolved issues with CLI test failures by using typer.echo for output and fixing indentation errors.

Release v0.2.0

Changelog

All notable changes to this project will be documented in this file.

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

[0.2.0] - 2025-11-03

Added

• Storage Abstraction Layer: Introduced a flexible storage system with a BaseStorage abstract base class.
• JSON and SQLite Storage: Implemented JsonStorage (default) and SqliteStorage for persisting agent execution
  traces.
• CLI Trace Explorer: Added a new agenthelm traces command group to the CLI.
• traces list command: List recent traces with pagination and JSON output option.
• traces show command: Show detailed information for a specific trace.
• traces filter command: Filter traces by various criteria like tool name, status, date, execution time, and
  confidence score.
• traces export command: Export filtered traces to JSON, CSV, or Markdown formats.
• Observability Documentation: Added docs/observability.md with detailed information on the new features.
• Observability Example: Added a new example in examples/observability_example/ to demonstrate the new storage and
  CLI features.

Changed

• Standardized Logging: Replaced all print() statements with Python's standard logging module for better control
  over output verbosity.
• CLI Entry Point: The main CLI entry point is now orchestrator.cli:app.
• Dependencies: Added tabulate as a new dependency for table formatting in the CLI.
• Project Structure: Moved storage-related classes to the orchestrator/core/storage/ directory.

Fixed

• Corrected various import errors and fixed linter warnings identified by ruff.
• Fixed an issue where the SQLite database file was being deleted after table creation in the example agent.
• Resolved issues with CLI test failures by using typer.echo for output and fixing indentation errors.