Token Panel ULTIMATE

🎛️ Know your limits. Stay within them. Maximize your capacity.

Real-time usage tracking for Claude Max, Gemini, and Manus — all in one place.

---

Features

Provider	What It Tracks
Claude Max	5-hour window, 7-day window, reset times
ChatGPT / OpenAI	Per-model API rate limits (requests + tokens), Plus subscription caps
Gemini	RPD/RPM/TPM per model, bottleneck detection
Manus	Daily refresh, monthly credits, addon balance

Plus a webchat widget that shows it all at a glance.

---

Claude Max Usage

Track your Claude Max subscription usage in real-time.

What It Shows

• 5-hour window: Rolling usage percentage and reset time
• 7-day window: Weekly usage percentage and reset time
• Model-specific limits: Sonnet and Opus allocations

Usage

# Pretty print current usage
python3 {baseDir}/scripts/claude-usage-fetch.py

# Update JSON file for the widget
python3 {baseDir}/scripts/claude-usage-fetch.py --update

# Raw JSON output
python3 {baseDir}/scripts/claude-usage-fetch.py --json

Requirements

• Claude Code CLI installed and authenticated (claude /login)

Auto-Update (Optional)

# Add to crontab for automatic updates every 5 minutes
*/5 * * * * python3 {baseDir}/scripts/claude-usage-fetch.py --update

---

Gemini Multi-Model Tracking

Track the bottleneck metric (highest % among RPD, RPM, TPM) for each model.

Model Limits (Tier 1)

Model	RPM	TPM	RPD
gemini-3-pro	25	1M	250
gemini-2.5-pro	25	1M	250
gemini-2.5-flash	2000	4M	∞
gemini-3-flash	1000	1M	10K
gemini-2.0-flash	2000	4M	∞

Fallback Strategy

gemini-3-pro → gemini-2.5-pro → gemini-2.5-flash(∞) → gemini-3-flash → gemini-2.0-flash(∞)

Most capable first, unlimited RPD models as safety nets.

Reset time: Midnight Pacific (RPD resets daily)

JSON Format

Store in memory/gemini-usage.json:

{
  "models": {
    "gemini-3-pro": {
      "limits": { "rpm": 25, "tpm": 1000000, "rpd": 250 },
      "usage": { "rpm": 17, "tpm": 1380000, "rpd": 251 },
      "status": "exceeded"
    }
  }
}

---

Manus Credit Monitoring

Credit Structure

• Monthly: 4,000 credits (resets on renewal)
• Daily refresh: 300 credits (resets 01:00)
• Addon: Purchased credits (never expire)

Usage

# Pretty print current usage
python3 {baseDir}/scripts/manus-usage-fetch.py

# Update JSON file for the widget
python3 {baseDir}/scripts/manus-usage-fetch.py --update

# Raw JSON output
python3 {baseDir}/scripts/manus-usage-fetch.py --json

Requirements

• MANUS_API_KEY environment variable set

Auto-Update (Optional)

# Add to crontab for automatic updates every 15 minutes
*/15 * * * * MANUS_API_KEY=your-key python3 {baseDir}/scripts/manus-usage-fetch.py --update

JSON Format

Store in memory/manus-usage.json (auto-generated by fetch script):

{
  "credits": {
    "total_all_time": 8407,
    "breakdown": {
      "monthly": { "used": 480, "limit": 4000, "remaining": 3520 },
      "addon": 7296
    },
    "daily_refresh": { "used": 0, "remaining": 300, "limit": 300, "reset_time": "01:00 local" }
  },
  "today": { "tasks": 0, "credits_used": 0, "breakdown": [] }
}

---

ChatGPT / OpenAI Usage

Track API rate limits per model and ChatGPT Plus subscription caps.

What It Shows

• API rate limits: Requests/min, tokens/min per model (from response headers)
• Plus caps: GPT-4o (150/3hr), o3 (100/week), o4-mini (300/day), etc.
• Models probed: gpt-4o, gpt-4o-mini, gpt-4, gpt-3.5-turbo, o3-mini

Usage

# Pretty print current usage
python3 {baseDir}/scripts/chatgpt-usage-fetch.py

# Update JSON file for the widget
python3 {baseDir}/scripts/chatgpt-usage-fetch.py --update

# Raw JSON output
python3 {baseDir}/scripts/chatgpt-usage-fetch.py --json

Requirements

• OPENAI_API_KEY environment variable set
• Key needs chat completion permissions (rate limit headers are extracted from minimal probe calls)
• For billing/cost data: key needs api.usage.read scope (admin key)

Auto-Update (Optional)

# Add to crontab for automatic updates every 10 minutes
*/10 * * * * OPENAI_API_KEY=your-key python3 {baseDir}/scripts/chatgpt-usage-fetch.py --update

---

Budget-Aware Behavior

Add to your SOUL.md:

## Resource Awareness

**Behavior by budget level:**
| Budget | Behavior |
|--------|----------|
| 🟢 >50% | Normal operations |
| 🟡 30-50% | Be concise |
| 🟠 10-30% | Defer non-essential tasks |
| 🔴 <10% | Minimal responses only |

Agent Self-Check

import json
from pathlib import Path

def get_claude_usage():
    path = Path.home() / ".openclaw/workspace/memory/claude-usage.json"
    if path.exists():
        data = json.loads(path.read_text())
        return data.get("limits", {}).get("five_hour", {}).get("utilization", 0)
    return 0

---

Webchat Widget

A Tampermonkey userscript that displays real-time usage in OpenClaw webchat.

Installation

1. Install Tampermonkey

Browser	Link
Chrome	Chrome Web Store
Firefox	Firefox Add-ons
Edge	Edge Add-ons
Safari	Mac App Store

2. Create New Script

1. Click Tampermonkey icon → "Create a new script..."
2. Delete all default content
3. Copy entire contents of {baseDir}/scripts/budget-panel-widget.user.js
4. Paste into Tampermonkey
5. Ctrl+S to save

3. Refresh Webchat

Go to http://localhost:18789 and refresh. Panel appears bottom-left.

Troubleshooting

• Panel not appearing? Check Tampermonkey is enabled
• Shows 0%? Run claude-usage-fetch.py --update first
• MIME error? Full restart: openclaw gateway stop && openclaw gateway start

---

Files

token-panel-ultimate/
├── SKILL.md
├── package.json
└── scripts/
    ├── claude-usage-fetch.py       # Claude Max usage fetcher
    ├── manus-usage-fetch.py        # Manus credit usage fetcher
    └── budget-panel-widget.user.js # Webchat widget

---

Gateway Plugin

For full integration, the budget-panel gateway plugin is available in our OpenClaw fork:

Repository: github.com/globalcaos/clawdbot-moltbot-openclaw

The plugin provides:

• budget.usage gateway method for real-time data
• Automatic JSON file reading
• Multi-provider aggregation

Install the plugin at extensions/budget-panel/ in your OpenClaw installation.

---

Related Skills

• shell-security-ultimate - Command security enforcement
• agent-memory-ultimate - Memory system with usage logs

---

Credits

Created by Oscar Serra with the help of Claude (Anthropic).

Built during a late-night hacking session, February 2026.