OpenClaw ships with no default token limits, which means every conversation can use unlimited tokens and rack up unexpected API costs. This guide walks you through setting up global, per-conversation, and per-skill token budgets to control spending while maintaining a smooth user experience.
Why This Is Hard to Do Yourself
These are the common pitfalls that trip people up.
No default limits
OpenClaw ships with no token limits. Every conversation can use unlimited tokens.
Unclear budget sizing
How many tokens does a typical conversation use? What's a reasonable daily limit? No guidance exists.
Hard vs soft limits
Hard limits cut off mid-response. Soft limits warn but don't stop. Neither is obvious to configure.
Step-by-Step Guide
Understand token costs
Learn what you're paying for.
# Rough token costs (per 1M tokens, as of 2025):
# Claude Opus: $15 input / $75 output
# Claude Sonnet: $3 input / $15 output
# Claude Haiku: $0.25 input / $1.25 output
#
# A typical conversation: 2,000-10,000 tokens
# A complex task: 20,000-100,000 tokensCreate the budgets config file
# Create config/budgets.yaml
mkdir -p ~/.openclaw/config
touch ~/.openclaw/config/budgets.yamlConfigure global budgets
# In config/budgets.yaml:
global:
daily_limit: 500000 # ~$7.50/day with Sonnet
monthly_limit: 10000000 # ~$150/month with Sonnet
warning_threshold: 0.8 # Warn at 80% of limit
hard_limit_action: pause # "pause" or "reject"Configure per-conversation limits
# In config/budgets.yaml (continued):
per_conversation:
max_tokens: 50000
warning_at: 40000
on_limit: warn # "warn", "pause", or "end"
message: "This conversation is approaching its token limit."Warning: Setting per-conversation limits too low (under 10,000) will cause conversations to feel cut short. Start with 50,000 and adjust based on your usage patterns.
Verify budgets are active
# Restart OpenClaw to load new config
npm restart
# Check budget status
curl http://localhost:3000/api/budgets
# Should return current usage and limitsNeed Help Sizing Your Budgets?
Getting token budgets right requires understanding your usage patterns. Our experts analyze your actual token consumption and configure optimal limits — not too tight, not too loose.
Get matched with a specialist who can help.
Sign Up for Expert Help →