diff --git a/scripts/token-burn.py b/scripts/token-burn.py new file mode 100755 index 0000000..9f77eaa --- /dev/null +++ b/scripts/token-burn.py @@ -0,0 +1,306 @@ +#!/usr/bin/env python3 +""" +Claude Code token burn rate analyzer. + +Scans ~/.claude/projects/*/*.jsonl session files and aggregates token usage +by day, session, model, and project. Outputs CSV or JSON for dashboarding. + +Usage: + python3 token-burn.py # last 7 days, table + python3 token-burn.py --days 30 # last 30 days + python3 token-burn.py --csv # CSV to stdout + python3 token-burn.py --json # JSON to stdout + python3 token-burn.py --by session # per-session breakdown + python3 token-burn.py --by project # per-project breakdown + python3 token-burn.py --by model # per-model breakdown + python3 token-burn.py --by hour # hourly heatmap +""" + +import argparse +import csv +import glob +import json +import os +import sys +from collections import defaultdict +from datetime import datetime, timedelta, timezone + +CLAUDE_DIR = os.path.expanduser("~/.claude/projects") +SESSIONS_DIR = os.path.expanduser("~/.claude/sessions") + +# Pricing per 1M tokens (as of 2026-03, Opus 4.6 / Sonnet 4.6) +PRICING = { + "claude-opus-4-6": {"input": 5.0, "output": 25.0, "cache_write": 10.0, "cache_read": 0.50}, + "claude-sonnet-4-6": {"input": 3.0, "output": 15.0, "cache_write": 6.0, "cache_read": 0.30}, + "claude-sonnet-4-5-20250929": {"input": 3.0, "output": 15.0, "cache_write": 6.0, "cache_read": 0.30}, + "claude-haiku-4-5-20251001": {"input": 1.0, "output": 5.0, "cache_write": 2.0, "cache_read": 0.10}, + "MiniMax-M2.7": {"input": 0.30, "output": 1.20, "cache_write": 0.375, "cache_read": 0.06}, +} +DEFAULT_PRICING = {"input": 5.0, "output": 25.0, "cache_write": 10.0, "cache_read": 0.50} + + +def estimate_cost(model, usage): + """Estimate USD cost from usage dict.""" + p = PRICING.get(model, DEFAULT_PRICING) + input_tok = usage.get("input_tokens", 0) + output_tok = usage.get("output_tokens", 0) + cache_write = usage.get("cache_creation_input_tokens", 0) + cache_read = usage.get("cache_read_input_tokens", 0) + cost = ( + input_tok * p["input"] / 1_000_000 + + output_tok * p["output"] / 1_000_000 + + cache_write * p["cache_write"] / 1_000_000 + + cache_read * p["cache_read"] / 1_000_000 + ) + return cost + + +def load_session_meta(): + """Load session metadata (name, cwd) from sessions dir.""" + meta = {} + for f in glob.glob(f"{SESSIONS_DIR}/*.json"): + try: + with open(f) as fh: + d = json.load(fh) + meta[d.get("sessionId", "")] = { + "name": d.get("name", ""), + "cwd": d.get("cwd", ""), + "started": d.get("startedAt", 0), + } + except (json.JSONDecodeError, KeyError): + pass + return meta + + +def scan_sessions(since_date): + """Scan all session JSONL files, yield per-message token records.""" + jsonl_files = glob.glob(f"{CLAUDE_DIR}/**/*.jsonl", recursive=True) + + # Filter by modification time for speed + cutoff_ts = since_date.timestamp() + jsonl_files = [f for f in jsonl_files if os.path.getmtime(f) >= cutoff_ts] + + for filepath in jsonl_files: + project = os.path.basename(os.path.dirname(filepath)) + session_id = os.path.splitext(os.path.basename(filepath))[0] + + try: + with open(filepath) as fh: + for line in fh: + try: + obj = json.loads(line) + except json.JSONDecodeError: + continue + + if obj.get("type") != "assistant": + continue + + msg = obj.get("message", {}) + usage = msg.get("usage") + if not usage: + continue + + ts_str = obj.get("timestamp", "") + if not ts_str: + continue + + try: + ts = datetime.fromisoformat(ts_str.replace("Z", "+00:00")) + except (ValueError, AttributeError): + continue + + if ts.date() < since_date.date(): + continue + + model = msg.get("model", "unknown") + input_tok = usage.get("input_tokens", 0) + output_tok = usage.get("output_tokens", 0) + cache_write = usage.get("cache_creation_input_tokens", 0) + cache_read = usage.get("cache_read_input_tokens", 0) + total = input_tok + output_tok + cache_write + cache_read + cost = estimate_cost(model, usage) + + yield { + "timestamp": ts, + "date": ts.strftime("%Y-%m-%d"), + "hour": ts.hour, + "session_id": session_id, + "project": project, + "model": model, + "input_tokens": input_tok, + "output_tokens": output_tok, + "cache_write_tokens": cache_write, + "cache_read_tokens": cache_read, + "total_tokens": total, + "cost_usd": cost, + } + except (IOError, OSError): + continue + + +def aggregate(records, group_by="date"): + """Aggregate records by the given key.""" + buckets = defaultdict(lambda: { + "input_tokens": 0, + "output_tokens": 0, + "cache_write_tokens": 0, + "cache_read_tokens": 0, + "total_tokens": 0, + "cost_usd": 0.0, + "messages": 0, + "sessions": set(), + }) + + for r in records: + if group_by == "hour": + key = f"{r['hour']:02d}:00" + else: + key = r[group_by] + + b = buckets[key] + b["input_tokens"] += r["input_tokens"] + b["output_tokens"] += r["output_tokens"] + b["cache_write_tokens"] += r["cache_write_tokens"] + b["cache_read_tokens"] += r["cache_read_tokens"] + b["total_tokens"] += r["total_tokens"] + b["cost_usd"] += r["cost_usd"] + b["messages"] += 1 + b["sessions"].add(r["session_id"]) + + # Convert sets to counts + result = {} + for key, b in sorted(buckets.items()): + b["session_count"] = len(b.pop("sessions")) + result[key] = b + + return result + + +def format_tokens(n): + if n >= 1_000_000: + return f"{n / 1_000_000:.1f}M" + if n >= 1_000: + return f"{n / 1_000:.1f}K" + return str(n) + + +def print_table(agg, group_label="Date"): + """Print a readable ASCII table.""" + print(f"\n{'─' * 90}") + print(f" {group_label:<20} {'Input':>8} {'Output':>8} {'CacheW':>8} {'CacheR':>8} {'Total':>9} {'Cost':>8} {'Msgs':>5}") + print(f"{'─' * 90}") + + grand = defaultdict(float) + grand["messages"] = 0 + + for key, b in agg.items(): + label = key[:20] if len(str(key)) > 20 else key + print( + f" {label:<20} " + f"{format_tokens(b['input_tokens']):>8} " + f"{format_tokens(b['output_tokens']):>8} " + f"{format_tokens(b['cache_write_tokens']):>8} " + f"{format_tokens(b['cache_read_tokens']):>8} " + f"{format_tokens(b['total_tokens']):>9} " + f"${b['cost_usd']:>6.2f} " + f"{b['messages']:>5}" + ) + for k in ["input_tokens", "output_tokens", "cache_write_tokens", "cache_read_tokens", "total_tokens", "cost_usd"]: + grand[k] += b[k] + grand["messages"] += b["messages"] + + print(f"{'─' * 90}") + print( + f" {'TOTAL':<20} " + f"{format_tokens(int(grand['input_tokens'])):>8} " + f"{format_tokens(int(grand['output_tokens'])):>8} " + f"{format_tokens(int(grand['cache_write_tokens'])):>8} " + f"{format_tokens(int(grand['cache_read_tokens'])):>8} " + f"{format_tokens(int(grand['total_tokens'])):>9} " + f"${grand['cost_usd']:>6.2f} " + f"{int(grand['messages']):>5}" + ) + print(f"{'─' * 90}\n") + + # Burn rate summary + days = len(agg) + if days > 0: + avg_day = grand["cost_usd"] / days + avg_tok = int(grand["total_tokens"]) / days + print(f" Avg/day: {format_tokens(int(avg_tok))} tokens | ${avg_day:.2f}") + print(f" Projected/month: {format_tokens(int(avg_tok * 30))} tokens | ${avg_day * 30:.2f}") + print() + + +def print_csv(agg, group_label="date"): + """Print CSV to stdout.""" + writer = csv.writer(sys.stdout) + writer.writerow([ + group_label, "input_tokens", "output_tokens", "cache_write_tokens", + "cache_read_tokens", "total_tokens", "cost_usd", "messages", "session_count" + ]) + for key, b in agg.items(): + writer.writerow([ + key, b["input_tokens"], b["output_tokens"], b["cache_write_tokens"], + b["cache_read_tokens"], b["total_tokens"], f"{b['cost_usd']:.4f}", + b["messages"], b["session_count"] + ]) + + +def print_json(agg, group_label="date"): + """Print JSON to stdout.""" + out = [] + for key, b in agg.items(): + row = {group_label: key} + row.update(b) + row["cost_usd"] = round(row["cost_usd"], 4) + out.append(row) + json.dump(out, sys.stdout, indent=2, default=str) + print() + + +def main(): + parser = argparse.ArgumentParser(description="Claude Code token burn rate analyzer") + parser.add_argument("--days", type=int, default=7, help="Look back N days (default: 7)") + parser.add_argument("--since", type=str, help="Start date YYYY-MM-DD (overrides --days)") + parser.add_argument("--by", choices=["date", "session", "project", "model", "hour"], + default="date", help="Group by (default: date)") + parser.add_argument("--csv", action="store_true", help="Output CSV") + parser.add_argument("--json", action="store_true", help="Output JSON") + args = parser.parse_args() + + if args.since: + since = datetime.strptime(args.since, "%Y-%m-%d").replace(tzinfo=timezone.utc) + else: + since = datetime.now(timezone.utc) - timedelta(days=args.days) + + group_map = { + "date": "date", + "session": "session_id", + "project": "project", + "model": "model", + "hour": "hour", + } + + sys.stderr.write(f"Scanning sessions since {since.strftime('%Y-%m-%d')}...\n") + records = list(scan_sessions(since)) + sys.stderr.write(f"Found {len(records)} assistant messages with token data\n") + + if not records: + sys.stderr.write("No data found.\n") + return + + agg = aggregate(records, group_by=group_map[args.by]) + + label_map = {"date": "Date", "session": "Session", "project": "Project", "model": "Model", "hour": "Hour"} + + if args.csv: + print_csv(agg, args.by) + elif args.json: + print_json(agg, args.by) + else: + print_table(agg, label_map[args.by]) + + +if __name__ == "__main__": + main() diff --git a/token-burn-dashboard.html b/token-burn-dashboard.html new file mode 100644 index 0000000..f328d59 --- /dev/null +++ b/token-burn-dashboard.html @@ -0,0 +1,798 @@ + + + + + +Token Burn Dashboard + + + +
+
+

Token Burn Dashboard

+ +

View

+
+ + + +
+ +

Machines

+
+ + + Local (macOS) +
+
+ + + openclaw-vm +
+ +

Models

+
+ +

Token Types

+
+ + + +
+ +

Chart Style

+
+ + + +
+ +

Scale

+
+ + +
+ +

Pricing ($/MTok)

+
+
+ Cache write = 1hr tier. Edit to adjust. +
+
+ +
+
+ +
+ + + + + +
+ +
+
+
+ +
+
+ +
+
+
+ +
+
+ +
+
+
+ +
+
+ +
+
+

Cost Breakdown by Model + Token Type

+
+
+
+ +
+
+

Token Type Distribution

+ +
+
+

Cache Efficiency (Read / Total Cache)

+ +
+
+
+
+ + + +