Helicone rate limits
Helicone rate limits throttle requests before they reach the provider — stopping runaway agent loops without waiting for the provider bill.
Hub: Tokenminning with Helicone.
This page answers helicone rate limit, helicone throttle, and helicone spending cap.
Why rate limit at the gateway
Agent loops can send hundreds of requests per hour. Provider dashboards show the damage after the fact. Helicone rate limits return HTTP 429 before the provider call — saving tokens and money.
Configure limits
In the Helicone dashboard → Rate Limits:
| Limit type | Example | Use |
|---|---|---|
| Requests per minute | 60 RPM | Interactive dev |
| Requests per day | 500 RPD | CI eval scripts |
| Tokens per day | 1M TPD | Production feature cap |
Attach limits to API keys, users, or custom properties.
Per-environment patterns
| Environment | Rate limit |
|---|---|
| CI / eval | 10 RPM, haiku-only model |
| Dev | 60 RPM |
| Prod | Higher cap with alerting |
Pair with cheap model defaults — rate limits are the backstop, routing is the first lever.
Agent loop protection
// In your agent middleware — respect 429 from Helicone
if (response.status === 429) {
throw new Error("Rate limit hit — stop agent loop");
}Combine Helicone rate limits with maxSteps in your agent framework.
Spending guardrails
- Rate-limit CI keys aggressively (10–30 RPM)
- Alert when daily token limit reaches 80%
- Separate limits per feature via custom properties
- Never run unbounded agent loops without a gateway throttle
Troubleshooting
429 in production — limit too low for legitimate traffic. Raise cap or split features into separate keys.
Limit not enforcing — verify requests route through Helicone (check Requests tab for the call).