Rudel: Open-Source Analytics for Claude Code

Rudel is an open-source analytics dashboard that tracks and visualizes metrics for Claude Code sessions. By using a CLI hook, it automatically captures token usage, session duration, and full transcripts for analysis. The platform is available as a hosted service or can be self-hosted for maximum data control.

Key Points

• Provides a dedicated dashboard for tracking Claude Code metrics including token usage and session duration.
• Utilizes a CLI-based hook system to automate the upload of session transcripts and metadata.
• Collects detailed data such as Git context, project paths, and full prompt/response content.
• Offers both a free hosted service and the ability to self-host using Docker and ClickHouse.
• Maintains an open-source MIT license, encouraging community contribution and transparency.

Sentiment

The community is generally receptive to the concept of AI coding session analytics, recognizing it as a real unmet need. However, there is meaningful skepticism about the privacy implications of the default cloud-hosted approach and questions about whether the tool is over-engineered for its purpose. The responsive engagement from the Rudel team helps maintain a constructive tone overall.

In Agreement

• Analytics for AI coding sessions fills a genuine need, as teams have no visibility into session efficiency or patterns
• The open-source, self-hostable approach is the right model for a tool handling sensitive session data
• The finding that skills are underutilized (4% usage rate) resonates with developer experience and highlights a real gap in AI tooling
• Cross-team analytics and session sharing can help developers learn from each other's patterns and avoid repeated mistakes
• The session abandonment and error cascade findings provide actionable insights for improving AI-assisted workflows

Opposed

• Uploading Claude Code session transcripts to a third party is a significant privacy and security risk, even with self-hosting available
• The infrastructure is over-engineered — ClickHouse and Docker for what amounts to a small dataset that could be analyzed with simple text files
• Claude Code's built-in /insights command already provides similar functionality without the setup overhead
• The dataset of 1,573 sessions from a single team of 6 is too small and homogeneous to draw generalizable conclusions
• Session abandonment may be misinterpreted — short sessions are often intentional for focused, single-purpose tasks