Aimogen ProDocumentation
Product page Support

API and provider errors

Decoding the error messages Aimogen Pro surfaces from AI providers, what causes each, and how to resolve them.

Provider errors appear verbatim in System & Logs › Activity Logs. Find yours below.

Aimogen Pro classifies every error into one of nine categories, which decides whether it is retried, whether the model is cooled down, and whether the key is held. See Errors, retries and timeouts.

Authentication#

Contains: invalid_api_key, incorrect api key, invalid api key, authentication_error, invalid x-api-key, api_key_invalid, unauthorized, HTTP 401

Symptoms#

Every request fails immediately. Nothing generates anywhere.

Likely causes#

A wrong or revoked key; a copied newline or trailing space; a key from a different organisation or project; the key in the wrong provider field.

Resolution#

  1. Settings › API Keys, re-copy the key, and use Test API key.
  2. Confirm it is in the right field — an Anthropic key in the OpenAI field fails this way.
  3. Check the key still exists in the provider dashboard.
  4. For OpenAI, clear the Organization ID field unless you specifically need it.

How to diagnose#

Test API key performs a real minimal request and shows the provider response verbatim.

Quota and billing#

Contains: insufficient_quota, exceeded your current quota, insufficient funds, insufficient credits, out of credits, payment required, billing

Symptoms#

Requests fail with a billing-related message, often after working previously.

Likely causes#

No credit on the provider account; a spend cap reached; a prepaid balance exhausted; an expired payment method.

Resolution#

Top up or fix billing at the provider. A ChatGPT Plus subscription does not include API credit — that is a separate purchase. Anthropic and xAI are prepaid.

How to diagnose#

The provider usage or billing dashboard is authoritative. System & Logs › Usage Logs shows what you consumed.

Rate limits#

Contains: rate limit, too many requests, resource_exhausted, requests per minute, tokens per minute, HTTP 429

Symptoms#

Requests succeed individually but fail during bulk runs, or fail in bursts.

Likely causes#

Your account tier limits requests or tokens per minute; too many parallel requests; a free tier with tight limits.

Resolution#

  1. Raise Delay Between API & Scraping Requests (ms), and the image equivalent.
  2. Add more keys to the provider field, one per line.
  3. Enable Reliability Mode with smart key rotation, which honours Retry-After and rotates to a healthy key.
  4. Reduce posts per run.
  5. Raise your account tier with the provider.

How to diagnose#

Rate limits correlate with volume. If it happens only during bulk runs, this is it.

Model errors#

Contains: model_not_found, model not found, no such model, unknown model, has been deprecated, no longer available, invalid model, no endpoints found

Symptoms#

One rule, chatbot or agent fails while others work. Often starts immediately after a plugin update.

Likely causes#

The model was retired by the provider and removed in a plugin release; your account does not have access to it; an OpenRouter route is unavailable.

Resolution#

Open the affected rule, chatbot or agent and select a current model. See Selecting models.

Some models require organisation verification or a spend tier before the provider enables them.

How to diagnose#

The error names the model. Check whether it still appears in the selector.

Context length#

Contains: context_length_exceeded, maximum context length, context window, prompt is too long, request_too_large, too many tokens

Symptoms#

Long articles or long conversations fail; short ones work.

Likely causes#

Prompt plus expected output exceeds the model window; too much retrieval context injected; too much conversation history; internet access adding large scraped content.

Resolution#

  1. Reduce the retrieval chunk count and raise the score threshold.
  2. Reduce Max Chat Messages To Send As API Context.
  3. Lower the context limit settings under Advanced AI Settings so prompts are truncated earlier.
  4. Generate in sections rather than one call.
  5. Use a larger-context model.

Note that this category is not retried on another model — an oversized prompt stays oversized.

How to diagnose#

Tokens, context and output length.

Content filters and safety#

Contains: content_filter, content policy, safety filter, prohibited_content, responsible ai, was blocked by, finish_reason":"safety, safety ratings

Symptoms#

Empty output, sometimes with no error at all. Certain topics fail consistently.

Likely causes#

The provider blocked the prompt or the response. Azure applies stricter filters than OpenAI directly. Google returns an empty response with a SAFETY finish reason rather than an error.

Resolution#

Rewrite the prompt. Topics that trip filters most often: weapons, health claims, alcohol, adult content, violence, and anything phrased as impersonation.

This category is deliberately not retried on another model — a blocked prompt is usually blocked everywhere.

How to diagnose#

Test the same prompt in the Playground, simplifying until it passes. That isolates the offending phrase.

Permission#

Contains: permission_error, permission denied, does not have access, not authorized to access, HTTP 403

Symptoms#

Authentication succeeds but the specific request is refused.

Likely causes#

An OpenAI Organization ID pointing at the wrong organisation; a restricted key without access to that model or endpoint; an Azure deployment the key cannot use.

Resolution#

Clear or correct the Organization ID; check key restrictions at the provider; confirm the Azure deployment belongs to the same resource as the key.

Provider outages#

Contains: overloaded, api_error, internal server error, service unavailable, temporarily unavailable, bad gateway, gateway timeout, HTTP 500/502/503/504, 529

Symptoms#

Intermittent failures with no pattern in your configuration.

Likely causes#

The provider is shedding load or having an incident. Anthropic 529 and OpenAI 503 are the common ones.

Resolution#

This is what Reliability Mode exists for. With it enabled and a cross-provider fallback list, an outage is survived rather than suffered. Two such failures also trip the provider circuit breaker, so the plugin stops trying that provider for the cooldown period.

How to diagnose#

Check the provider status page. If your fallback provider works, it is not you.

Network#

Contains: could not resolve host, failed to connect, connection refused, connection timed out, network is unreachable, ssl connect error

Symptoms#

Every provider fails, or one specific provider fails while others work.

Likely causes#

Outbound HTTP blocked by the host or a firewall; DNS failure; a security plugin proxying wp_remote_*; the provider domain not on an egress allowlist.

Resolution#

  1. System & Logs › System Info — check Remote POST and Remote GET.
  2. Ask your host whether outbound HTTPS to the provider domain is permitted.
  3. Test with security plugins deactivated.
  4. See Plugin and server conflicts.

Timeouts#

Symptoms#

Long generations die at a consistent number of seconds.

Likely causes#

PHP max_execution_time, the plugin request timeout, or the provider timeout — whichever is smallest.

Resolution#

Work out which limit that number corresponds to, and raise it. See Errors, retries and timeouts and Recommended server configuration.

Reasoning models and sonar-deep-research are inherently slow; a non-reasoning model may be the better answer.

When the error is not shown#

If a feature fails with no error:

  1. Enable Enable Logging and Enable Detailed Logging under Settings › General Settings.
  2. Reproduce while watching System & Logs › Live AI Activity.
  3. Check Usage Logs — no row means no request was made.
  4. Check the browser console for a failing admin-ajax.php request.

Still stuck? Open a support ticket and include the diagnostics from Aimogen Pro › System & Logs › System Info.