Skip to content

wooo-jin/clinsight

Repository files navigation

Clinsight

Automatically record, analyze, and compound your Claude Code sessions

Quick Start · Install Guide · Features · How It Works · 한국어

Node Platform License Tests


Why Clinsight?

If you use Claude Code, you've probably experienced this:

"What did I discuss with Claude yesterday..." "Session got lost, now I'm explaining everything from scratch" "That prompt was so good... which session was it again?" "Didn't realize I spent $30+ in a single session until it was too late"

Sessions are ephemeral, but the insights from them don't have to be.

Clinsight automatically records and analyzes your Claude Code usage, turning today's trial-and-error into tomorrow's assets.


Quick Start

npm (recommended)

npm install -g clinsight

# Register hooks with Claude Code (starts automatic session recording)
clinsight-setup

# Launch the dashboard
clinsight

From source

git clone https://github.com/wooo-jin/clinsight.git
cd clinsight
pnpm install && pnpm build

# Register hooks with Claude Code
pnpm setup

# Launch the dashboard
pnpm dev

See INSTALL.md for detailed installation instructions.


Features

1 — Automatic Session Archive

Hooks into Claude Code's hook system to automatically record every conversation. Recording starts when a session begins and creates a complete archive when it ends. No manual steps — just use Claude Code as usual.

2 — TUI Dashboard

A 7-tab terminal dashboard accessible right from your terminal:

+-----------------------------------------------------------+
|  Clinsight  | 42 sessions | updated 17:30                 |
|                                                           |
|  Dashboard  Insights  Sessions  Cost  Compound  Archive   |
|                                                           |
|  +-- Today --+  +-- Week ---+  +-- Month --+             |
|  | Sessions: 8|  | Sessions: 34|  | Sessions: 142|       |
|  | Cost: $4.20|  | Cost: $18.5 |  | Cost: $72.30 |       |
|  | Score: 82  |  | Score: 78   |  | Score: 75    |       |
|  +-----------+  +------------+  +-------------+          |
|                                                           |
|  7-day cost trend    ............                          |
|  7-day efficiency    ............                          |
|                                                           |
|  [1-7] tabs  [Tab] navigate  [r] refresh  [q] quit       |
+-----------------------------------------------------------+
Tab Description
Dashboard Today/weekly/monthly summary, cost trends, efficiency scores
Insights Churn detection, context saturation warnings, improvement suggestions
Sessions Per-session analysis (tokens, tool usage, edited files)
Cost Cost breakdown by model and project
Compound AI-powered pattern extraction and knowledge compounding
Archive Browse full session conversation history
Settings Configure archive retention periods

3 — Session Analysis Engine

Each session is automatically analyzed and assigned an efficiency score (0-100).

Efficiency = f(first-try rate, churn index, exploration efficiency, context saturation, duration)
Metric How It's Measured
Churn Index Number of edit reversions (not just repeated edits)
Exploration Efficiency Ratio of read-only files that didn't contribute to edits
Context Saturation Actual usage vs. model context window size
First-Try Rate Percentage of edits completed without reversions
Cost Tracking Based on actual per-token pricing by model

4 — Compound (Knowledge Compounding)

A nightly cron job sends the day's sessions to Claude for deep analysis.

Today's sessions  -->  Pattern recognition   -->  Actionable rules
                       Solution extraction        for CLAUDE.md
                       Convention mining
                       Golden prompt preservation

Today's struggles become tomorrow's assets.

5 — Real-Time Alerts

Analyzes the current session on every prompt. When issues are detected, warnings are injected into Claude's context so Claude naturally addresses them in its responses.

Condition What Claude Says
Context 75%+ "You may need /compact soon"
Context 90%+ "Running /compact is recommended"
Cost $10+ "Costs are getting high"
3+ reversions "Consider re-evaluating your approach"
60min+ session "40-50 minute sessions are more efficient"

No separate notification popups. Claude mentions it naturally in conversation.

6 — Zombie Session Detection

Detects orphaned processes and abandoned session directories, with one-click cleanup.


How It Works

Claude Code Session
       |
       +-- SessionStart --> Hook --> Initialize archive
       |
       +-- PromptSubmit --> Hook --> Sync conversation + real-time analysis
       |                              |
       |                              +--> If warning threshold met:
       |                                    Inject into Claude's context
       |
       +-- SessionStop  --> Hook --> Create complete archive
                                         |
                                    +----+-----+
                                    |          |
                              TUI Dashboard  Cron (23:00)
                              Live browsing  Compound analysis
                                             Pattern/solution extraction

Tech Stack

Runtime Node.js 18+
UI Ink + React (Terminal UI)
Language TypeScript strict mode
Architecture Feature-Sliced Design (FSD)
Integration Claude Code Hooks
Test Vitest (109 tests)
Platform macOS, Linux, Windows

Acknowledgements

Inspired by claude-session-tracker.


License

MIT

About

No description, website, or topics provided.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors