0.1.5

New Features

• Single Agent Mode: Introduce Agent.run_stream() for running a single super-agent with tool-call decision making, without requiring a team. Supports multi-turn conversation with persistent context across calls.
• Long-term Memory: Add memory module with chunking, embedding, vector search, and automatic summarization. Agents can now store and retrieve long-term memories across sessions via MemoryManager.
• Memory Tools: Add memory_search and memory_get built-in tools for agents to actively query memory during execution.
• Built-in Tools Expansion: Add a comprehensive set of built-in tools:
  • bash — Execute shell commands with security checks and output truncation
  • read — Read file contents with support for images and partial reads
  • write — Write files with automatic directory creation
  • edit — Patch files by exact text replacement with diff output
  • grep — Search file contents using ripgrep
  • find — Search files by glob pattern
  • ls — List directory contents
• Web UI: Add frontend framework (Vue.js) for visual interaction with agent teams.
• WebSocket Support: Add WebSocket communication framework for real-time multi-agent streaming output.
• Workspace Configuration: Add global workspace management (agentmesh.set_workspace()) for organizing agent data and memory files.

Bug Fixes

• Fix max_turns None check in stream executor to prevent runtime errors with enhanced error logging.
• Fix TypeError caused by missing None check in _trim_messages context management.
• Fix memory_manager being required in MemorySearchTool and MemoryGetTool; it is now optional.
• Fix bash tool to better balance security restrictions and practical usability.
• Fix requests dependency version to resolve SDK installation failure under uv.