How Much Does OpenClaw Cost to Run?

# OpenClaw is MIT-licensed open source software:
# https://github.com/openclaw/openclaw

# Cost breakdown:
✅ OpenClaw software: $0
✅ Installation: $0
✅ Updates and patches: $0
✅ No per-user fees: $0
✅ No subscription: $0

# You pay for:
❌ Hosting (if cloud-based)
❌ API provider fees (Anthropic, OpenAI, etc.)
❌ Your time to set up and maintain it

# Option 1: Local (your machine)
# Cost: $0/month (electricity negligible)
# Pros: Free, full control, no data leaves your network
# Cons: Only available when your machine is on, no remote access

# Option 2: Cloud VPS (DigitalOcean, AWS EC2, etc.)
# Cost: $12-50/month for basic setup
# - DigitalOcean Droplet (2 vCPU, 4GB RAM): ~$24/month
# - AWS t3.medium (2 vCPU, 4GB RAM): ~$30/month
# - Hetzner CX21 (2 vCPU, 4GB RAM): ~$6/month
# Pros: Always on, remote access, scalable
# Cons: Ongoing monthly fee

# Option 3: GPU-enabled for local models (optional)
# Cost: $300-1000/month if you want to run local LLMs
# - AWS p3.2xlarge (1 V100 GPU): ~$3/hour (~$2200/month if always on)
# - RunPod H100 GPU instance: ~$2.50/hour (~$1800/month)
# Pros: Run models locally, no per-token API fees
# Cons: Very expensive, complex setup, often slower than APIs

# Recommendation for most teams:
# Start local for testing ($0), move to cloud VPS for production ($20-30/month).

# API pricing (as of early 2025):
# Anthropic Claude 3.5 Sonnet:
# - Input: $3 per 1M tokens
# - Output: $15 per 1M tokens

# OpenAI GPT-4o:
# - Input: $2.50 per 1M tokens
# - Output: $10 per 1M tokens

# Example cost per conversation:
# - Typical conversation: 5,000 input tokens, 1,500 output tokens
# - Claude Sonnet cost: (5000 × $3/1M) + (1500 × $15/1M) = $0.015 + $0.0225 = $0.0375 per conversation
# - ~27 conversations per $1
# - 1,000 conversations/month = ~$37.50/month

# Heavy usage example (power user):
# - 100 conversations/day × 30 days = 3,000 conversations/month
# - 3,000 × $0.0375 = $112.50/month per power user

# Team of 10 with mixed usage:
# - 5 light users (100 conversations/month each): $18.75
# - 3 medium users (500 conversations/month each): $56.25
# - 2 power users (3,000 conversations/month each): $225
# Total: ~$300/month in API costs

# Simple per-conversation cost formula:
per_conversation_cost = (avg_input_tokens × input_price_per_token) + (avg_output_tokens × output_price_per_token)

# Example with Claude Sonnet:
# - Average input: 5,000 tokens
# - Average output: 1,500 tokens
# - Input price: $3 per 1M tokens = $0.000003 per token
# - Output price: $15 per 1M tokens = $0.000015 per token

# Calculation:
(5000 × 0.000003) + (1500 × 0.000015) = $0.015 + $0.0225 = $0.0375 per conversation

# To find your actual average:
# 1. Run OpenClaw for a week
# 2. Check provider dashboard for total tokens used
# 3. Divide by number of conversations
# 4. Use that average for future budgeting

# Anthropic Console (console.anthropic.com):
# Settings → Billing → Usage Alerts
# - Set alert at $50, $100, $200, etc.
# - Email notifications when threshold crossed

# OpenAI Dashboard (platform.openai.com):
# Settings → Limits → Usage Limits
# - Set hard limit (stops API calls when reached)
# - Set soft limit (sends email warning)

# CloudWatch (if using AWS):
# Create billing alarm:
aws cloudwatch put-metric-alarm \
  --alarm-name "OpenClaw-API-Cost-Alert" \
  --alarm-description "Alert when API costs exceed $100" \
  --metric-name EstimatedCharges \
  --namespace AWS/Billing \
  --statistic Maximum \
  --period 21600 \
  --evaluation-periods 1 \
  --threshold 100 \
  --comparison-operator GreaterThanThreshold

# Budget optimization strategies:

# 1. Use cheaper models for simple tasks:
# - Route simple queries to Claude Haiku ($0.25/$1.25 per 1M tokens)
# - Reserve Sonnet/Opus for complex reasoning

# 2. Set per-user token budgets:
# - Daily limit: 50,000 tokens per user
# - Monthly limit: 1M tokens per user
# - Prevents one power user from burning entire budget

# 3. Cache repeated content:
# - Use Anthropic's prompt caching (90% cost reduction for cached tokens)
# - Cache system prompts, context documents, etc.

# 4. Compress context:
# - Summarize long documents before feeding to LLM
# - Remove redundant information from prompts

# 5. Monitor and adjust:
# - Review usage weekly
# - Identify expensive queries
# - Optimize or restrict them

# Example monthly budget allocation (team of 10, $200/month total):
# - Hosting (cloud VPS): $30
# - API costs: $170
# - Per-user limit: $17/month (170 / 10 users)
# - Daily per-user: ~450 conversations at $0.0375 each