💰 SmartMeter

AI cost optimization for OpenClaw — Analyze usage patterns, generate optimized configs, and cut costs by 48%+ without sacrificing quality.

What It Does

SmartMeter helps you save money on AI API costs by:

•Analyzing your OpenClaw usage patterns from session logs
•Identifying tasks that use expensive models unnecessarily
•Classifying workloads (code, writing, research, config, etc.)
•Generating optimized openclaw.json configurations
•Recommending specialized agents for different task types
•Monitoring budget usage with alerts

Real-World Results

Tested on live OpenClaw data:

•Current cost: $59.97/month
•Optimized cost: $31.14/month
•Savings: $28.82/month (48.1%)

The key insight: 69% of tasks worked perfectly with cheaper models, while premium models were only needed for 15% of complex work.

Installation

Via ClawHub (Recommended)

bash

npx clawhub@latest install smartmeter

Via npm

bash

npm install -g openclaw-smartmeter

Manual Installation

bash

git clone https://github.com/vajih/openclaw-smartmeter.git
cd openclaw-smartmeter
npm install
npm link

Quick Start

One Command - Dashboard Included! 🚀

bash

smartmeter analyze

This will:

•✅ Analyze your OpenClaw usage logs
•✅ Calculate potential savings
•✅ Generate recommendations
•✅ Launch interactive dashboard automatically
•✅ Show real-time cost breakdown

The dashboard opens at http://localhost:8080 with:

•Cost savings visualization
•Model usage breakdown
•One-click optimization
•Export reports

Usage

Basic Workflow

bash

# 1. Analyze usage (opens dashboard)
smartmeter analyze [session-file.jsonl]

# 2. Preview optimizations (without applying)
smartmeter preview

# 3. Apply optimizations to openclaw.json
smartmeter apply

# 4. Rollback if needed
smartmeter rollback

Advanced Commands

bash

# Quick cost evaluation
smartmeter evaluate examples/sample-session.jsonl

# Open dashboard (if already analyzed)
smartmeter dashboard

# Export analysis as JSON
smartmeter export --format json --output analysis.json

# Show current version
smartmeter --version

# Get help
smartmeter --help

Features

📊 Usage Analysis

•Parse JSONL session logs from ~/.openclaw/agents/*/sessions/
•Extract model usage, token counts, costs, and cache performance
•Track usage across all agents and time periods
•Minimum 2 days recommended, 14+ days for best results

🎯 Task Classification

•Automatically categorize tasks: code, writing, research, config, other
•Keyword-based classification using prompts and tool usage
•Identify patterns in AI agent behavior
•Success rate tracking by model and category

💡 Cost Optimization

•Identify expensive models used for simple tasks
•Recommend cheaper alternatives with same quality
•Calculate potential savings with confidence levels
•Generate specialized agents for different workloads

⚙️ Config Generation

•Production-ready openclaw.json generation
•Primary model optimization based on analysis
•Specialized agent creation (code-reviewer, researcher, quick-tasks)
•Budget controls and spending alerts
•Cache optimization settings
•Fallback chains for reliability

🎨 Interactive Dashboard

•Real-time cost breakdown and projections
•Visual charts for model usage and savings
•One-click optimization application
•Export analysis reports
•OpenRouter API integration
•Responsive design (mobile, tablet, desktop)

🔐 Safe Deployment

•Automatic config backups before changes
•Timestamped backup files
•One-command rollback: smartmeter rollback
•Config validation before applying
•Merge with existing configs (preserves custom settings)

🔌 OpenRouter Integration

•Connect your OpenRouter API key
•View actual usage vs. projected
•Real-time budget monitoring
•Cost alerts via Telegram (optional)

Configuration

OpenRouter API Key (Optional)

For live usage monitoring and alerts:

bash

# Via environment variable
export OPENROUTER_API_KEY="or-v1-xxxxxxxxxxxx"

# Or store in OpenClaw config
smartmeter config --key or-v1-xxxxxxxxxxxx

Dashboard Configuration

The dashboard server runs on:

•Dashboard: http://localhost:8080
•API Server: http://localhost:3001

To change ports, edit ~/.openclaw/smartmeter/config.json

How It Works

1. Data Collection

SmartMeter reads your OpenClaw session logs:

code

~/.openclaw/agents/*/sessions/*.jsonl

Each log entry contains:

•Model used
•Tokens (input, output, cache)
•Cost breakdown
•User prompts
•Tool invocations
•Timestamps

2. Analysis Engine

•Parser: Extracts usage data from JSONL logs
•Classifier: Categorizes tasks by type (code, writing, research, etc.)
•Aggregator: Calculates costs, averages, and patterns
•Recommender: Generates optimization suggestions

3. Optimization Logic

For each task category:

•Analyze current model usage
•Calculate success rates
•Compare costs vs. quality
•Recommend optimal model
•Estimate savings

Example:

code

Category: code
Current: Claude Opus 4.5 ($0.80/task)
Success rate: 94%

Optimal: Claude Sonnet 4.5 ($0.38/task)
Success rate: 94% (same quality)

Savings: $0.42/task × 200 tasks = $84/month (52.5%)

4. Config Generation

Creates optimized openclaw.json:

json

{
  "agents": {
    "defaults": {
      "model": {
        "primary": "openrouter/auto",
        "fallback": ["claude-sonnet-4-5", "claude-opus-4-5"]
      },
      "budget": {
        "daily": 5.0,
        "weekly": 30.0,
        "alert": {
          "telegram": true,
          "threshold": 0.75
        }
      }
    },
    "code-reviewer": {
      "model": "claude-sonnet-4-5",
      "description": "Handles code review, debugging, and programming tasks",
      "budget": { "daily": 2.0 }
    },
    "quick-tasks": {
      "model": "deepseek/chat",
      "description": "Simple queries, quick answers, config changes",
      "budget": { "daily": 0.5 }
    }
  }
}

Dashboard Guide

Cost Overview

•Current monthly projection
•Optimized projection
•Potential savings ($ and %)
•Confidence level

Model Breakdown

•Usage by model
•Token counts
•Cost per model
•Success rates

Recommendations

•Actionable optimization suggestions
•Expected savings per recommendation
•One-click implementation
•Risk level indicators

Charts & Visualizations

•Cost trends over time
•Model usage distribution
•Task category breakdown
•Cache hit rate analysis

Actions

•Preview: See changes before applying
•Apply: Implement optimizations (creates backup)
•Export: Download analysis as JSON/CSV
•Refresh: Update with latest session data

Best Practices

Data Requirements

•Minimum: 2 days of usage data
•Recommended: 14+ days for accurate patterns
•Optimal: 30+ days for seasonal variations

Analysis Timing

Run analysis:

•After major project milestones
•Monthly for budget reviews
•When costs seem high
•Before scaling usage

Safety Tips

•Always preview before applying
•Keep backups (automatic)
•Start with conservative settings
•Monitor for 1-2 days after changes
•Roll back if quality degrades

Optimization Strategy

•Phase 1: Replace expensive models for simple tasks
•Phase 2: Create specialized agents
•Phase 3: Optimize caching and budgets
•Phase 4: Fine-tune based on results

Troubleshooting

Dashboard Not Loading

bash

# Check if servers are running
ps aux | grep node

# Restart servers
killall node
smartmeter analyze

No Session Data Found

bash

# Check if session logs exist
ls -la ~/.openclaw/agents/*/sessions/

# Use sample data for testing
smartmeter analyze examples/sample-session.jsonl

OpenRouter API Errors

bash

# Verify API key
echo $OPENROUTER_API_KEY

# Test connection
curl -H "Authorization: Bearer $OPENROUTER_API_KEY" \
  https://openrouter.ai/api/v1/auth/key

Apply Failed

bash

# Check config backup
ls -la ~/.openclaw/openclaw.json.backup-*

# Rollback to previous
smartmeter rollback

# Validate config manually
cat ~/.openclaw/openclaw.json | jq '.'

Command Reference

`smartmeter analyze [file]`

Analyze OpenClaw usage and launch dashboard.

Options:

•[file] - Optional path to specific session file
•Without file: analyzes all sessions in ~/.openclaw/agents/

Output:

•Analysis saved to ~/.openclaw/smartmeter/analysis.json
•Dashboard opens at http://localhost:8080
•API server starts on port 3001

`smartmeter preview`

Preview optimizations without applying.

Shows:

•Proposed config changes
•Expected savings
•Risk assessment
•Confidence levels

`smartmeter apply`

Apply optimizations to openclaw.json.

Safety:

•Creates backup: openclaw.json.backup-{timestamp}
•Validates config before applying
•Merges with existing settings
•Preserves custom configurations

`smartmeter rollback`

Restore previous configuration.

Process:

•Lists available backups
•Restores most recent by default
•Verifies restoration success

`smartmeter dashboard`

Open dashboard without re-analyzing.

Requirements:

•Previous analysis must exist
•Servers start automatically

`smartmeter evaluate <file>`

Quick cost evaluation of session file.

Use Case:

•One-off analysis
•Testing with sample data
•CI/CD cost checks

`smartmeter export`

Export analysis data.

Options:

•--format json|csv - Output format
•--output <file> - Destination file

`smartmeter config`

Manage SmartMeter configuration.

Subcommands:

•--key <api-key> - Store OpenRouter API key
•--show - Display current config
•--reset - Reset to defaults

Examples

Basic Usage

bash

# Install
npm install -g openclaw-smartmeter

# Analyze and view dashboard
smartmeter analyze

# Apply optimizations
smartmeter apply

Advanced Workflow

bash

# Analyze specific session
smartmeter analyze ~/.openclaw/agents/my-agent/sessions/session-123.jsonl

# Preview changes
smartmeter preview

# Export analysis
smartmeter export --format json --output report.json

# Apply optimizations
smartmeter apply

# Monitor for issues, rollback if needed
smartmeter rollback

Continuous Monitoring

bash

# Weekly analysis (add to cron)
0 0 * * 0 smartmeter analyze && smartmeter export --format csv --output weekly-report.csv

# Budget alerts via Telegram
export TELEGRAM_BOT_TOKEN="your-token"
export TELEGRAM_CHAT_ID="your-chat-id"
smartmeter monitor --alert-threshold 0.75

FAQ

Will this affect my AI quality?

No. SmartMeter only recommends changes when cheaper models achieve the same success rate. You can always preview before applying.

How much data is needed?

Minimum 2 days, but 14+ days recommended for accurate patterns. The tool will warn if data is insufficient.

Is my data safe?

Yes. All analysis happens locally on your machine. No data is sent to external servers except OpenRouter API (if you configure it).

Can I undo changes?

Yes. Every change creates an automatic backup. Use smartmeter rollback to restore.

Does it work with other AI providers?

Currently optimized for OpenRouter, but works with any provider supported by OpenClaw.

What if I use custom agents?

SmartMeter merges with your existing config and preserves custom settings. It only modifies cost-related fields.

Requirements

•Node.js: 18.0.0 or higher
•OpenClaw: Any version with session logging
•Storage: At least 2 days of session logs
•Ports: 8080 (dashboard), 3001 (API)

Learn More

•GitHub: https://github.com/vajih/openclaw-smartmeter
•Issues: https://github.com/vajih/openclaw-smartmeter/issues
•License: Apache 2.0

Support

Found a bug? Have a feature request?

•Check existing issues
•Create a new issue with details
•Include your Node.js version and OS

Changelog

v0.3.0 (February 11, 2026)

•✨ OpenRouter API integration
•🎨 Professional UI redesign
•🐛 Fixed apply optimization feature
•📊 Enhanced dashboard with real-time data
•🔒 Improved config validation

v0.2.4

•💄 UX improvements for cost display
•🐛 Zero-cost scenario handling

v0.2.3

•🐛 Critical bug fix: keep analyze command alive

v0.2.2

•📦 Initial public release
•✨ Core analysis and optimization engine
•🎨 Basic dashboard UI

Made with 💰 by the OpenClaw community