Moraine: Unified Agent Tracing (github.com)

Moraine has emerged as a significant tool for the AI/ML community, enabling local trace indexing of agent sessions from various harnesses like Codex, Claude Code, and Pi Coding Agent into ClickHouse. This unified local stack offers a comprehensive monitor UI and exposes Multi-Context Protocol (MCP) retrieval capabilities, allowing agents to maintain a searchable long-term memory across different tools and sessions. By creating a single, indexed record of conversations and interactions, users can effectively track and manage the activities of multiple AI agents simultaneously. The development of Moraine emphasizes real-time insights and keyword-based search capabilities rather than semantic search, enhancing the efficiency of users in querying their agents. Key features include adaptive ingestion adapters for integrating session traces, detailed guidance on querying past interactions, and the ability to peek into live agent sessions. With ongoing updates and active development, Moraine presents a promising advancement in the way developers and researchers can oversee, control, and refine agent operations across various frameworks, ultimately streamlining workflows in AI-driven tasks.