Rudel: Open-source analytics for Claude Code sessions
Rudel is an open-source analytics layer for Claude Code sessions that provides visibility into how developers actually use AI coding agents. The team built it after realizing they had no insight into their own Claude Code usage patterns.
Key findings from 1,573 real sessions
The team analyzed 1,573 Claude Code sessions containing 15M+ tokens and 270K+ interactions. Some notable findings:
- Skills were only being used in 4% of sessions
- 26% of sessions are abandoned, most within the first 60 seconds
- Session success rate varies significantly by task type (documentation scores highest, refactoring lowest)
- Error cascade patterns appear in the first 2 minutes and predict abandonment with reasonable accuracy
- There is no meaningful benchmark for 'good' agentic session performance
How Rudel works
The tool consists of a CLI and dashboard. To get started:
npm install -g rudel
rudel login # authenticate via your browser
rudel enable # auto-upload sessions when Claude Code exitsFor batch upload of existing sessions:
rudel upload # interactive picker for batch uploadTechnical implementation
When you run rudel enable, it registers a Claude Code hook that runs when a session ends. The hook uploads the session transcript to Rudel. Transcripts are stored in ClickHouse and processed into analytics.
Data collected
Each uploaded session includes:
- Session ID & timestamps (start, last interaction)
- User ID & organization ID
- Project path & package name
- Git context (repository, branch, SHA, remote)
- Session transcript (full prompt & response content)
- Sub-agent usage
Security considerations
Rudel ingests full coding-agent session data, which may contain sensitive material including source code, prompts, tool output, file contents, command output, URLs, and secrets. The team emphasizes only enabling Rudel on projects and environments where you're comfortable uploading that data. For the hosted service at app.rudel.ai, they state they do not have access to personal data contained in uploaded transcripts.
Development and self-hosting
The project is MIT licensed and includes documentation for local development and self-hosting. See CONTRIBUTING.md for local setup and docs/self-hosting.md for running your own instance.
This type of tool is particularly useful for teams wanting to understand how their developers interact with AI coding assistants, identify patterns in successful vs. abandoned sessions, and potentially improve their workflows based on data rather than anecdote.
📖 Read the full source: HN AI Agents
👀 See Also
depct: MCP Server Provides Claude with Live Runtime Analysis and Documentation
depct is an MCP server that instruments Node.js applications to capture runtime data, generating structured documentation with confidence levels that Claude can access before coding. The tool updates documentation automatically after Claude makes changes.
OpenYak: Open-Source Desktop AI Agent for Local File Management and Automation
OpenYak is an open-source desktop AI assistant that runs entirely on your machine, offering file management, data analysis, and office automation with 100+ AI models through OpenRouter and 20+ BYOK providers.
Building syntaqlite: A SQLite DevTools Project Created with AI Assistance
Lalit Maganti built syntaqlite, a set of developer tools for SQLite, over three months using AI coding agents after wanting it for eight years. The project required parsing SQL exactly like SQLite, which involves adapting SQLite's dense C codebase with over 400 grammar rules.
Skill Bill: A Markdown-Based Governance Framework for AI Coding Skills
A developer built Skill Bill, a framework of 44 Markdown-based AI skills for Kotlin, Android/KMP, PHP, and Go that addresses prompt management problems like naming drift and duplicated logic. It includes orchestrator skills like 'feature-implement' that chain 10-12 skill invocations and syncs to Claude Code, Copilot, GLM, and Codex.