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.
Unreleased¶
0.1.7 - 2026-02-03¶
Added¶
- Add comprehensive display rendering system for agent messages:
MessageRendererABC withRichRenderer,PlainTextRenderer, andHtmlRendererimplementationsDisplayPresetsystem withCOMPACT,DETAILED, andVERBOSEpresets (accessible viaget_preset())- Standalone
print_stats(),print_timeline(),print_tools()functions for standalone message visualization MessageQuery.print_stats(),print_timeline(),print_tools()convenience methods- Rich
__rich_console__protocol support for direct Rich console rendering of message analytics
Fixed¶
- Fix ruff formatting issues and remove unused GitHub Actions workflows
0.1.6 - 2026-02-02¶
Added¶
- Add
MessageQueryclass for comprehensive message filtering, analytics, and export: - Filter messages by role, tool_name, content (with regex support), and slice operations
- Analytics methods:
stats()(message counts and token statistics),tool_summary()(tool call analytics),timeline()(conversation turns) - Multi-format export: JSON, Markdown, CSV, and dict formats
Agent.messagesproperty returnsMessageQueryfor stateless, on-demand access to conversation history- Works with or without context tracking and token counting
- Add
MessageStats,ToolCallInfo, andTurndata models for message analytics
0.1.5 - 2026-02-02¶
Fixed¶
- Fix graceful error handling not being applied to tools passed directly to
Agent(..., tools=[...]); constructor tools are now wrapped with error handler whengraceful_tool_errorsis enabled (default behavior)
0.1.4 - 2026-01-26¶
Added¶
- Add markdown template support with YAML frontmatter as alternative to Jinja2 templates
- Add
TemplateTypeenum to distinguish between JINJA2 and MARKDOWN template types - Add
MarkdownParseErrorfor markdown template parsing failures - Add
TemplateConflictErrorwhen both.mdand.jinja2templates exist for same name - Add
file_extensionsconfig option (replacesfile_extensionwith backward compatibility) - Add
examples/directory with runnable scripts demonstrating all major features - Add enhanced module docstrings with usage examples and cross-references
Fixed¶
- Fix MCP
test_connectionusing incorrect snake_case attributeinput_schemainstead of camelCaseinputSchema
Changed¶
- Update documentation with v0.1.3 MCP features (Streamable HTTP transport, connection testing)
- Update README and MCP user guide with new transport types and connection testing API
- Update
config.example.tomlwith[prompts]section and Streamable HTTP MCP example
0.1.3 - 2026-01-22¶
Added¶
- Add Streamable HTTP transport support for MCP servers (auto-detected from URL when not ending in
/sse) - Add connection testing methods to
MCPClientManager:test_connection(),test_all_connections(), and sync variants - Add
get_server()method toMCPClientManagerfor retrieving individual server instances - Add
MCPToolInfoandMCPConnectionResultmodels for connection testing results - Add new MCP errors:
MCPConnectionError,MCPConnectionTimeoutError,MCPServerNotFoundError
Fixed¶
- Fix deprecated
get_event_loop().run_until_complete()usage in MCP client (now usesasyncio.run())
0.1.2 - 2026-01-22¶
Added¶
- Add graceful tool error handling that converts tool exceptions to
ModelRetry, allowing the LLM to receive error messages and attempt recovery instead of crashing the agent loop - Add
graceful_tool_errorsconfig option toAgentConfig(default:True) - Add
graceful_errorsparameter totool()andtool_plain()decorators for per-tool override
0.1.1 - 2026-01-21¶
Changed¶
- Simplify codebase by consolidating duplicate patterns across agent, workflows, and compaction strategies (net ~180 lines reduced)
Removed¶
- Remove unused placeholder configs
ObservabilityConfigandStreamingConfigfrommamba_agents.configsubmodule
Fixed¶
- Fix incorrect MCP API documentation in README (
get_toolsets()->as_toolsets())
0.1.0 - 2025-01-20¶
Added¶
- Initial release of Mamba Agents framework
- Core
Agentclass wrapping pydantic-ai with built-in context and token tracking - Configuration system with
AgentSettingssupporting environment variables,.envfiles, and TOML/YAML - Context window management with 5 compaction strategies:
- sliding_window
- summarize_older
- selective_pruning
- importance_scoring
- hybrid
- Token counting and cost estimation with
TokenCounter,UsageTracker,CostEstimator - Prompt template management with Jinja2 support via
PromptManagerandPromptTemplate - MCP (Model Context Protocol) integration:
MCPClientManagerfor managing MCP servers- Support for stdio and SSE transports
- File-based config loading from
.mcp.jsonfiles - Workflow orchestration framework:
- Base
Workflowclass for custom implementations - Built-in
ReActWorkflow(Thought-Action-Observation loop) - Configurable hooks for workflow events
- Built-in tools for filesystem, glob, grep, and bash operations
- Model backends for Ollama and vLLM (OpenAI-compatible)
- Comprehensive error handling with retry logic and circuit breaker pattern
- Observability with logging and optional OpenTelemetry tracing