FAQ

Common errors on the Agentsflare platform.

{"statusCode":400,"message":"Model gpt-image-2 is not available, please check if the model is in the admin allowed range"}

This error indicates that the API key you are using does not have permission to use the gpt-image-2 model. Please refer to Configure Model

{"statusCode":404,"message":"Key not found"}

This error indicates that the API key you are using does not exist. Please check if your API key is correct.

{"message":"Cannot POST /openai/v2/images/generations","error":"Not Found","statusCode":404}

This error indicates that the requested URL is incorrect. Please check if your request URL is correct.

What causes a 429 error?

A 429 Too Many Requests response means the request was rate-limited. Possible causes:

• Insufficient account balance: Your credits have been exhausted. Please top up your account and retry.
• Concurrency / rate limit exceeded: Too many requests were sent in a short period, exceeding the rate limit of your current channel tier. Please reduce request frequency or implement retry logic with exponential backoff.

What should I do when I get a 5xx error?

500, 502, 503, 529, and other 5xx errors are upstream model server errors, meaning Agentsflare forwarded your request correctly but the model provider returned an error. Common causes include:

• The upstream model service is temporarily overloaded or unavailable
• The upstream service is under maintenance or experiencing an outage

When encountering these errors, it is recommended to retry after a short delay, or implement automatic retry logic with exponential backoff. If the error persists, please report it through official support channels.

Is there a timeout limit?

Agentsflare itself has no timeout setting — requests will wait indefinitely until the model returns a result.

However, Agentsflare uses Cloudflare as a network-layer proxy, and Cloudflare enforces a 300-second (5-minute) idle timeout: if no data is transmitted between either end of the connection for 300 seconds, Cloudflare will automatically close the connection.

For potentially long-running requests, we recommend:

• Using streaming output (stream: true). In streaming mode, the model continuously sends tokens, keeping data flowing and effectively avoiding the 300-second idle disconnection.
• Keeping max_completion_tokens (max_tokens for older models) reasonable to avoid generating excessively long responses.

What does the IP_NOT_ALLOWED error mean?

Example:

{"error": {"type":"forbidden_error","message":"IP address not allowed: 155.*.*.4","code":"IP_NOT_ALLOWED"}}

This error means the request's source IP is not included in the IP Whitelist configured in the console, so the request was rejected. How to fix:

• Go to the console -> IP Whitelist, and add the IP address or the corresponding CIDR range.
• If you use proxies, NAT, load balancers, or cloud providers (AWS/GCP/Azure), make sure to add the egress/public IP(s) or CIDR ranges.
• To find your server's public IP, check https://api.ipify.org or a similar service from the machine making the request.

Reference: IP Whitelist