existentialburn

Extract your Claude Code usage stats. Runs 100% locally — your prompts and code never leave your machine. Only token counts, tool calls, timestamps, and model names are extracted.

Quick start

npx existentialburn > upload.json

Then upload at existentialburn.com/upload.

What it extracts

The extractor reads ~/.claude/projects/ JSONL conversation files and outputs structured JSON:

• Daily aggregates — tokens, cost, messages, sessions, tool calls, model breakdowns, hourly activity
• Session details — duration, token counts, models used, tool calls, git branch, subagent spawns
• Totals — lifetime token/cost/session/message/day counts
• Metadata — streaks, longest session, tool call breakdown, project count, date range

What it does NOT extract

• Prompt content
• Code snippets
• File contents
• Conversation text
• Any personally identifiable information beyond usage patterns

Programmatic usage

import { extract } from "existentialburn";

const data = extract();
console.log(`$${data.totals.totalCost.toFixed(2)} across ${data.totals.totalSessions} sessions`);

Custom directory

const data = extract({ claudeDir: "/path/to/claude/projects" });

Output format

{
  "version": 1,
  "extractedAt": "2026-03-15T...",
  "daily": [{ "date": "2026-03-14", "totalCost": 12.50, ... }],
  "sessions": [{ "id": "...", "durationMs": 3600000, ... }],
  "totals": { "totalCost": 450.00, "totalSessions": 200, ... },
  "meta": { "longestStreak": 14, "subagentSpawns": 42, ... }
}

Requirements

• Node.js 18+
• Claude Code installed (~/.claude/projects/ must exist)

License

MIT