We built our AI agent, for analyzing CI logs (www.mendral.com)

A new AI agent named Mendral has been developed to analyze Continuous Integration (CI) logs, demonstrating significant advantages over generalist models like Claude Code in diagnosing and fixing CI failures. Designed by a team familiar with CI systems, Mendral autonomously closes over 16,000 CI investigations each month, leveraging a specialized architecture that integrates a log ingestion pipeline processing billions of CI log lines weekly. Unlike Claude Code, Mendral has access to extensive contextual data, including historical run logs and failure patterns, enabling it to effectively correlate issues and identify root causes that would be cumbersome for human engineers to track. Mendral employs a multi-agent structure utilizing various models from Anthropic, each tailored for distinct tasks—such as evidence collection and complex reasoning—while ensuring efficient cost management and performance. Its robust architecture includes features like durable execution to handle transient failures gracefully and a lightweight isolation strategy for executing tests, allowing the agent to resume operations swiftly after interruptions. This comprehensive system not only enhances reliability and efficiency but also evolves continuously through real-world feedback, making Mendral a valuable tool for teams managing extensive CI pipelines, such as PostHog and Luminai.