Local analytics dashboard for Claude Code — with environmental impact tracking and mental health awareness.
Reads directly from ~/.claude/. No cloud, no API key, no telemetry.
The standard Claude Code dashboard tells you what you've done. This one also asks what it cost — in dollars, in energy, and in rest.
- 💻 Full technical depth — sessions, costs, tools, projects, memory, streaks.
- 🌱 Ecological footprint — LLM inference consumes real energy and water. This dashboard makes it visible, per model, per day, per token type.
- 🧠 Mental health — AI tools blur the line between focus and burnout. The disconnection panel tracks when you actually unplug.
git clone https://github.com/sebWarembourg/Lumen.git
cd Lumen
npm install
npm run dev # → http://localhost:3333To point at a different Claude Code profile:
CLAUDE_CONFIG_DIR=~/.claude-work npm run dev
|
|
Sessions, tokens, estimated cost, and local storage at a glance. Trend cards with sparklines, date presets (7d / 30d / 90d / custom range picker), model distribution, peak hours, project activity, and recent sessions.
Every token has a cost beyond dollars.
A dedicated Impact page estimates the energy (Wh), water (mL), and CO₂ (g) consumed by your Claude usage — broken down by model, day, and token type.
Methodology:
- Energy coefficients from Epoch AI research on LLM inference (input: 390 Wh/MTok · output: 5× at 1 950 Wh/MTok)
- PUE 1.14 (AWS datacenter overhead) · Water Usage Effectiveness: 1.8 mL/Wh
- Carbon intensity by grid: 🇺🇸 US (0.287 g CO₂/Wh) · 🇪🇺 EU (0.25) · 🇫🇷 France (0.06, nuclear)
- Model-size multipliers: Opus × 2.5 · Haiku × 0.4 · Sonnet × 1.0
Panels: hero KPIs (energy / water / CO₂), relatable equivalences (LED hours, EV km, flights Paris→NY), impact over time, impact by model, token-type breakdown donut, full factors table with sources.
From the "AI Brain Fry" talk: are you actually unplugging?
Inside the Activity page, the Disconnection block tracks the gaps — time spent without Claude.
| Metric | What it means |
|---|---|
| Break en cours | Consecutive days without Claude (0 = used today) |
| Repos / 30j | % of the last 30 days with no activity |
| Record | Longest disconnection streak ever |
A bar chart shows every gap in your history, color-coded:
- 🔴 1–2 days · 🟡 3–6 days · 🟢 7+ days (reference line for a real break)
For the activity : GitHub-style heatmap, current streak, longest streak, peak hours, day-of-week patterns.
Searchable, sortable project grid. Per-project: sessions, duration, estimated cost, languages, git branches, MCP/agent badges, top tools. Detail pages with cost over time, language distribution, branch activity.
Full session replay from JSONL. Badges for compaction, agents, MCP, extended thinking. Responses in GitHub-flavored Markdown. Tool calls and file cards inline. Per-turn cost and token breakdown.
Total cost, cache savings, cost without cache, cost over time, cost by project, per-model breakdown, cache efficiency panel.
Tool ranking, categories (file I/O, shell, agents, web, MCP…), feature adoption, error analysis, version history, git branch analytics.
History, Todos, Plans, Memory, and Settings — read directly from ~/.claude/. Give you visibility on the invisible.
- Global search:
Cmd+K/Ctrl+K// - Session navigation:
j/k·Enterto open ·Escto clear - Page shortcuts:
g sSessions ·g pProjects ·g cCosts - Responsive: desktop sidebar, mobile bottom nav, light/dark themes
| Path | Used for |
|---|---|
~/.claude/projects/<slug>/*.jsonl |
Session replay, tokens, tools |
~/.claude/stats-cache.json |
Aggregate stats (fast path) |
~/.claude/usage-data/session-meta/ |
Metadata fallback |
~/.claude/history.jsonl |
Command history |
~/.claude/todos/ · plans/ · memory/ |
Local files |
~/.claude/settings.json |
Settings, skills, MCP config |
Runs entirely locally. No login, no API key, no telemetry. Your data stays on your machine.
Originally derived from cc-lens by @Arindam200. Lumen — the standalone evolution — is built and maintained by @sebWarembourg, with Impact tracking, Disconnection panel, theme fixes, and UX improvements layered on top.