Skip to content
AnglinAI Docs

ERR_AI_OVERLOADED — AI Overloaded

HTTP Status: 503 Retryable: Yes Automatic retry: The callWithRetry utility retries up to 3 times with exponential backoff. This error surfaces only after all retries are exhausted.

What the User Sees

AI service is temporarily unavailable due to high demand. (ERR_AI_OVERLOADED)

What Causes This Error

Anthropic returns HTTP 529 (“overloaded”) when their infrastructure is under heavy load. This is fundamentally different from rate limiting (429) — it affects all customers globally, not just our API key. The 529 status code is Anthropic-specific and indicates their servers cannot accept new requests at all.

This typically occurs during periods of extremely high demand across the Anthropic platform, or during partial infrastructure incidents. The condition is usually temporary, resolving within minutes, but can last longer during major incidents.

The system automatically retries these requests via callWithRetry with exponential backoff. The user only sees this error if all 3 retry attempts fail, meaning the overload condition persisted for the full retry window (typically 30-60 seconds depending on backoff timing).

Unlike rate limiting, there is nothing the service can do to reduce the likelihood of this error — it is entirely dependent on the provider’s infrastructure health.

Resolution Steps

For Users

  1. Wait a few minutes and try again. This is a temporary condition with the AI provider.
  2. Try the “budget” quality tier, which uses Google Gemini instead of Anthropic. Gemini may not be affected by an Anthropic outage.
  3. Check https://status.anthropic.com to see if there is an ongoing incident.
  4. If the error persists for more than 15 minutes, consider trying again later.

For Administrators

  1. Check https://status.anthropic.com for active incidents or degraded performance notices.
  2. If the outage is prolonged (more than 30 minutes), consider temporarily routing all conversion traffic through Gemini by enabling budgetMode in the smart cascade converter configuration.
  3. Monitor the Grafana dashboard for the frequency of 529 errors. A spike that correlates with Anthropic’s status page confirms the root cause.
  4. No action is needed to “fix” this — the condition resolves when Anthropic’s infrastructure recovers.
  5. If users are reporting this frequently but Anthropic’s status page shows no issues, the problem may actually be rate limiting (ERR_AI_RATE_LIMITED) — check the raw HTTP status codes in the logs.