Skip to content

Showcase: claude-code-conversation-analyzer — analytics dashboard built on top of claude-code-log #130

@raedatoui

Description

@raedatoui

Hi @daaain — first off, huge thanks for claude-code-log. The SQLite cache it produces (claude-code-log-cache.db) turned out to be the perfect substrate for a project I've been building, and I wanted to share it here in case it's interesting to you or other users of this tool.

What I built

claude-code-conversation-analyzer — a dashboard that reads the cclog SQLite DB and layers on:

  • Cost analysis — per-session and per-model cost breakdowns using current Anthropic pricing (input/output/cache-creation/cache-read tokens)
  • Quality scoring — weighted scoring across cache usage, conversation length, query specificity, context utilization, and prompt efficiency, with letter grades (A–F)
  • Compaction detection — flags cache drops >10k tokens as compaction events
  • Better UI — Next.js 16 + React 19 dashboard with:
    • Project overview grid with grade distribution
    • Conversation list with sortable metrics
    • Expandable conversation cards with token charts (Recharts) and message transcripts
    • Full-text search across projects and sessions
    • Dark/light theme

How it integrates with claude-code-log

The pipeline runs claude-code-log as the data-generation step, then augments the resulting SQLite DB with two extra tables (session_costs, session_quality). Nothing in the cclog DB itself is modified — the analyzer is purely additive. A make all target handles the full flow: clone/update cclog fork → regenerate transcripts → run analysis → serve dashboard.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions