Ledger

Claude Code accounts. Your conversations, cost, and waste — in one private ledger.

Ledger is a local browser + terminal UI for every Claude Code conversation on your disk. It reads the JSONL sessions Claude Code writes to ~/.claude/projects/, and gives you a browser, a CLI, and a full Dashboard — token usage, cost, activity, waste patterns, model comparison, and git-correlated yield.

Everything runs on your machine. No proxy, no wrapper, no API keys. Nothing leaves your disk.

What you get

• Browser & search — every conversation you've ever had with Claude Code, with deep full-message search, project filters, bookmarks, and one-click resume.
• Dashboard — a CRED-style observability panel with five sub-tabs:
  • Overview — headline spend, cache hit rate, daily cost chart, 52-week activity heatmap, 13-category task classifier, top models, projects, tools, shell commands, MCP servers, and most-expensive sessions.
  • Optimize — a 6-pattern waste scanner with A–F health grade and copy-paste fixes (duplicate reads, low Read:Edit ratio, cache-creation overhead, junk directory reads, uncapped bash output, bloated CLAUDE.md).
  • Compare — side-by-side model performance: one-shot rate, retry rate, self-correction, cost per call / edit, output tokens / call, cache hit, delegation, planning, tools per turn.
  • Yield — correlates sessions with git log in each project and labels them productive, reverted, or abandoned.
  • Plan — subscription tracking for Claude Max / Pro / Cursor Pro / custom monthly budgets.
• Export — per-conversation .md / .json, zip of all conversations, full dashboard CSV/JSON.
• Terminal CLI — paginated interactive browser with keyboard navigation, search, bookmarks, and direct resume into Claude Code.
• Zero dependencies. Python standard library only.

Install

macOS / Linux

curl -fsSL https://raw.githubusercontent.com/AnshRajput/claude-chats-and-analytics-viewer/main/install.sh | sh

Windows (PowerShell)

iwr https://raw.githubusercontent.com/AnshRajput/claude-chats-and-analytics-viewer/main/install.ps1 | iex

Manual

pipx install claude-chats-and-analytics-viewer      # recommended
uvx claude-chats-and-analytics-viewer               # uv, no install
pip3 install --user claude-chats-and-analytics-viewer

Don't have pipx?

• macOS: brew install pipx && pipx ensurepath
• Linux: python3 -m pip install --user pipx && python3 -m pipx ensurepath
• Windows: pip install pipx && pipx ensurepath

Run a new terminal after ensurepath and type ccv.

Run

ccv                        # start Ledger (opens browser)
ccv --port 8080            # use a different port
ccv --no-open              # start without opening the browser
ccv --update               # update to latest
ccv --install              # auto-start on login (macOS LaunchAgent)
ccv --install-systemd      # auto-start on login (Linux systemd)
ccv --uninstall            # remove the auto-start service

Terminal CLI:

ccvc                              # interactive browser
ccvc -v                           # version
ccvc --check-update               # check PyPI for a newer version
ccvc --search "auth bug"          # filter by keyword
ccvc --project "myapp"            # filter by project
ccvc --view <session-id>          # view a conversation
ccvc --resume <session-id>        # resume in Claude Code
ccvc --list --limit 10            # non-interactive list

Keyboard shortcuts (CLI)

Dashboard API

The web UI is a thin layer over a local JSON API you can script against:

Every endpoint respects the selected period and stays 100% local.

How the task classifier works

Each assistant turn is assigned to one of 13 categories by matching tools used + keywords in your message (no LLM calls, deterministic):

From this the dashboard computes one-shot rate per category (the share of edit turns that didn't need a retry) — the cleanest signal for how well Claude got it right on the first try.

How it works

~/.claude/projects/<project-slug>/<session-id>.jsonl
                           │
                           ▼
  parser.py  →  per-turn data (tools, tokens, model, classifier)
                           │
                           ▼
  store.py   →  mtime-keyed metadata cache (v3, tmpdir)
                           │
                           ▼
  web.py  ←  HTTP server  →  single-page Dashboard UI
  cli.py  ←  terminal UI

Claude Code writes session transcripts as JSONL. Ledger reads them, parses per-turn usage + tool calls, caches the result, and serves everything from a local HTTP server on 127.0.0.1:5005. Nothing is uploaded.

Project structure

claude_conversation_viewer/
├── __init__.py              # __version__
├── web.py                   # HTTP handler + single-file HTML/CSS/JS UI
├── cli.py                   # Interactive terminal CLI
├── update_checker.py        # PyPI update check
├── pricing.py               # Model pricing table + cost helpers
├── classifier.py            # 13-category deterministic classifier
├── parser.py                # JSONL parsing (metadata + per-turn + full messages)
├── cache.py                 # Metadata cache v3, bookmarks, plan storage
├── store.py                 # ConversationStore
└── dashboard/
    ├── aggregator.py        # /api/dashboard payload
    ├── optimize.py          # Waste detectors + A–F grade
    ├── compare.py           # Model comparison
    ├── yield_tracker.py     # Git-correlated yield
    ├── plans.py             # Subscription presets
    ├── export.py            # CSV / JSON export
    └── period.py            # Period parsing

Troubleshooting

ccv not found after install Run pipx ensurepath and open a new terminal.

Port already in use Ledger automatically falls through to the next available port and prints it; or specify ccv --port 8080.

No conversations found Make sure you've run Claude Code at least once. Check:

ls ~/.claude/projects/
find ~/.claude/projects -name "*.jsonl" | head

Update check not working Requires network access to pypi.org. Fails silently by design.

Credits

• Dashboard observability inspired by codeburn — the terminal-first sibling that tracks the same data across 16 AI coding tools.
• Visual language: CRED-style premium minimalism — warm black neutrals, Inter Tight, JetBrains Mono, one chartreuse accent.

License

MIT — github.com/AnshRajput/claude-chats-and-analytics-viewer