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.