📊 OpenClaw Performance Profiler

Optimize costs and performance for OpenClaw workflows - Track token usage, identify bottlenecks, and reduce API costs.

🎯 Why This Matters

OpenClaw workflows can get expensive. This tool helps you:

• 💰 Track Costs - Monitor API usage and expenses
• ⚡ Find Bottlenecks - Identify slow operations
• 🎯 Optimize Tokens - Reduce unnecessary token consumption
• 📈 Visualize Metrics - Charts and reports
• 🔔 Set Budgets - Alert when costs exceed thresholds

🚀 Quick Start

Install

npm install -g openclaw-performance-profiler

Profile a Workflow

openclaw-profile start
# ... run your workflow ...
openclaw-profile stop
openclaw-profile report

Example Output

📊 OpenClaw Performance Profile

⏱️  Execution Time: 12.3s
💰 Total Cost: $0.45
🎯 Token Usage: 15,234 tokens

📈 Breakdown by Tool:

Tool            Calls  Time     Tokens   Cost
─────────────────────────────────────────────
web_fetch         12   2.4s     1,200    $0.02
web_search         3   1.8s     4,500    $0.09
exec              45   0.8s       450    $0.01
message            5   0.2s     9,084    $0.33
─────────────────────────────────────────────
Total             65   5.2s    15,234    $0.45

🔍 Optimization Suggestions:

⚠️  web_fetch is called 12 times - consider batching
💡 message uses 60% of tokens - simplify prompts
✅ exec is well-optimized

📖 Commands

openclaw-profile start [name]

Start profiling a session:

openclaw-profile start my-workflow

# Options:
#  --watch          Live monitoring
#  --threshold <n>  Alert when cost > threshold

openclaw-profile stop

Stop current profiling session:

openclaw-profile stop

openclaw-profile report [session]

Generate performance report:

openclaw-profile report

# Options:
#  --format <json|html|csv>
#  --output <file>
#  --compare <session>  Compare with another session

openclaw-profile analyze [file]

Analyze execution logs:

openclaw-profile analyze openclaw.log

# Detects:
# - Redundant API calls
# - Token waste
# - Slow operations
# - Cost spikes

openclaw-profile budget set <amount>

Set cost budget:

openclaw-profile budget set 10.00  # $10/day

# Options:
#  --period <day|week|month>
#  --alert <email|slack|webhook>

📊 Metrics Tracked

Token Usage

• Input tokens - Prompt/context sent to model
• Output tokens - Model response
• Total tokens - Sum of input + output
• Cost - Based on model pricing

Performance

• Execution time - Total duration
• Tool latency - Individual tool call times
• Queue time - Wait time for rate limits
• Throughput - Operations per second

Resource Usage

• API calls - Number of external requests
• Retries - Failed attempts that retried
• Cache hits/misses - If caching is enabled
• Memory - Peak memory usage

🎓 Examples

Example 1: Compare Two Sessions

openclaw-profile report session-1 --compare session-2

Output:

Comparing: session-1 vs session-2

Metric              Session 1    Session 2    Δ
─────────────────────────────────────────────────
Execution Time      12.3s        8.7s         -29% ✅
Total Cost          $0.45        $0.32        -29% ✅
Token Usage         15,234       10,891       -28% ✅
API Calls           65           48           -26% ✅

🎉 session-2 is 29% cheaper and 29% faster!

Example 2: Identify Token Waste

openclaw-profile analyze --focus tokens

Output:

🎯 Token Usage Analysis

Top Token Consumers:
1. message → 9,084 tokens (60%)
   - Prompt is 3.2KB per call
   - 💡 Suggestion: Reduce context window

2. web_search → 4,500 tokens (30%)
   - Returns full web pages
   - 💡 Suggestion: Use extractMode: 'text'

3. web_fetch → 1,200 tokens (8%)
   - ✅ Optimal

Potential Savings: 40% reduction → $0.18 saved

Example 3: Real-Time Monitoring

openclaw-profile start --watch

📡 Live Performance Monitor

Time   Tool         Tokens   Cost      Status
────────────────────────────────────────────────
12:01  web_fetch      120    $0.002    ✅
12:01  message      1,850    $0.037    ✅
12:02  web_search   1,500    $0.030    ✅
12:02  message      1,920    $0.038    ⚠️ High tokens
12:03  exec            50    $0.001    ✅

Current Total: $0.108 | Budget: $10.00 | 1% used

🔧 Integration

As Middleware

import { Profiler } from 'openclaw-performance-profiler';

const profiler = new Profiler({
  enabled: true,
  logLevel: 'info',
});

// Wrap your workflow
profiler.wrap(async () => {
  await myWorkflow();
});

const report = profiler.getReport();
console.log(report);

Programmatic API

import { Profiler } from 'openclaw-performance-profiler';

const profiler = new Profiler();

profiler.startSession('my-workflow');

// Track individual operations
profiler.track('web_fetch', async () => {
  return await tools.web_fetch({ url: 'https://example.com' });
});

profiler.endSession();

// Get insights
const insights = profiler.analyze();
console.log(insights.suggestions);

📈 Visualization

Generate Charts

openclaw-profile chart --type line --metric cost --output cost-over-time.png

Chart types:

• line - Trends over time
• bar - Comparison between tools/sessions
• pie - Proportion breakdown
• heatmap - Time-based activity

Dashboard (Bonus Feature)

Run local web dashboard:

openclaw-profile dashboard --port 8080

Features:

• Real-time metrics
• Historical trends
• Cost projections
• Alerts management

🎯 Optimization Strategies

1. Reduce Token Usage

Before:

await tools.message({
  message: `Here is a very long explanation with lots of unnecessary words 
            that could be simplified and made more concise...` // 50 tokens
});

After:

await tools.message({
  message: `Concise explanation.` // 5 tokens
});
// Saved: 45 tokens × $0.002/token = $0.09

2. Batch API Calls

Before:

for (const url of urls) {
  await tools.web_fetch({ url });  // 10 separate calls
}

After:

await Promise.all(
  urls.map(url => tools.web_fetch({ url }))
);  // Parallel execution, 60% faster

3. Cache Results

Before:

// Fetch same data multiple times
const data1 = await tools.web_fetch({ url });
const data2 = await tools.web_fetch({ url });  // Duplicate!

After:

const cache = new Map();
const cached = cache.get(url) || await tools.web_fetch({ url });
cache.set(url, cached);
// Saved: 1 API call

4. Use Cheaper Models

Some tasks don't need the most powerful model:

// For simple classification
profiler.suggest('use-cheaper-model', {
  current: 'claude-sonnet-4',
  recommended: 'claude-haiku-3',
  savings: '80%',
});

💰 Cost Models (Built-in)

| Model | Input ($/1K tokens) | Output ($/1K tokens) | |-------|---------------------|----------------------| | Claude Sonnet 4 | $0.003 | $0.015 | | Claude Haiku 3 | $0.00025 | $0.00125 | | GPT-4 Turbo | $0.010 | $0.030 | | GPT-3.5 Turbo | $0.0005 | $0.0015 |

Prices are estimates. Configure actual pricing in config.

🤝 Contributing

Contributions welcome! Ideas:

• [ ] More visualization types
• [ ] Integration with billing APIs
• [ ] ML-powered optimization suggestions
• [ ] Slack/Discord notifications
• [ ] Multi-tenant cost tracking

📄 License

MIT © Alex - Built for cost-effective OpenClaw

🔗 Links

• OpenClaw Pricing
• Token Optimization Guide
• Cost Calculator

Optimize performance. Reduce costs. Build smarter. 📊