OpenClaw Cost Control

OpenClaw users report spending $30-100+ per month on LLM API calls, with runaway loops causing bills of $3,600/mo or more. Without budget caps, a single agent loop can drain your API credits overnight.

Curate-Me’s gateway adds cost governance to every OpenClaw request with zero code changes.

How Cost Governance Works

Every LLM request passes through the gateway’s 15-stage governance chain. Step 2 is the cost estimation check:

Request → Rate Limit → COST CHECK → PII Scan → Model Allowlist → HITL Gate → Provider
                          ↓
                   Estimate tokens (tiktoken BPE)
                   Look up model pricing
                   Check per-request limit
                   Check daily budget
                   Check monthly budget
                          ↓
                   ALLOW or DENY

If the estimated cost would exceed any budget threshold, the request is denied before it reaches the provider. You never pay for a request that breaks your budget.

Budget Configuration

Per-Request Cost Limits

Set the maximum cost for a single LLM request:

# In the dashboard → Policies → Cost Controls
max_cost_per_request: 2.50  # USD — deny any request estimated > $2.50

Daily Budget Caps

Set a daily spending limit per organization:

daily_budget: 50.00  # USD — deny all requests once $50/day reached

Monthly Budget Caps

Set a monthly ceiling:

monthly_budget: 500.00  # USD — hard stop at $500/month

Per-Runner Session Budgets

For managed runners (private beta), set budgets per session:

session_budget: 5.00  # USD — terminate session when $5 spent

Cost Tracking

Real-Time Dashboard

The cost dashboard shows:

• Total spend today, this week, this month
• Cost by model — which models cost the most
• Cost by runner — which runners are most expensive
• Cost by channel — WhatsApp vs Telegram vs Slack
• Cost trend — daily spend over the last 30 days
• Projected monthly — based on current run rate

Cost Alerts

Configure alerts at budget thresholds:

Emergency Kill Switch

The kill switch immediately halts all agent activity across all runners. Use it when:

• A runaway loop is detected
• Costs spike unexpectedly
• A security incident requires immediate shutdown

Available from the dashboard header or via API:

curl -X POST https://api.curate-me.ai/v1/runners/kill-all \
  -H "X-CM-API-Key: cm_sk_xxx"

Model Allowlists

Prevent agents from using expensive models:

# Only allow these models
allowed_models: [
  "gpt-4o-mini",    # $0.15/1M input, $0.60/1M output
  "claude-haiku",    # $0.25/1M input, $1.25/1M output
  "deepseek-chat",   # $0.14/1M input, $0.28/1M output
]

Any request to a model not on the allowlist is denied.

Cost Optimization Tips

1. Set model allowlists — Block expensive models for non-critical agents
2. Use per-session budgets — Limit runaway risk per runner session
3. Enable cost alerts — Catch spikes before they become bills
4. Review cost-by-model weekly — Identify wasteful model usage
5. Set conservative daily budgets — Start at $25/day, increase as needed
6. Use the kill switch liberally — Better safe than sorry

Pricing Tiers

View full pricing →