multiplayer
About
Multiplayer attacks the weakest input to AI debugging: coding agents fix production bugs poorly because they see stack traces, not the unsampled full-stack session data — request and response bodies, headers, cross-service correlation — that APM tools sample away. The platform runs locally alongside a coding agent, captures complete session data through recorder SDKs spanning JavaScript, React Native, Go, .NET, Python, Ruby, and Java, and auto-correlates the traces across components. Its triage layer filters for high-priority bugs and deduplicates identical errors, so a recurring exception produces one merge-ready pull request rather than a flood of duplicates; the loop runs from error capture through agent prompting to PR creation and notification. The self-hostable stack (MongoDB, Kafka, ClickHouse, Redis, MinIO) deploys via Docker Compose, and a companion CLI provides a terminal surface for reviewing sessions and turning fixes into branches. Teams running Claude Code against production incidents are the target users; the project is young, with a small commit history and adoption still forming.