AI rate limit & provider errors

Rate-limit and “provider returned error” messages come from the AI provider, not from Atlarix itself. Common causes:

• The model's free tier or your provider quota is exhausted for now.
• The provider is temporarily overloaded (HTTP 429/503).
• The selected model isn't available to your account or region.
• A provider-specific request limit (tokens/minute, requests/minute) was hit.

1. 1Switch models: open the model selector under the chat input and pick a different model or provider — then press Retry.
2. 2Wait a moment: transient 429/503 errors usually clear within a minute. Retry the message.
3. 3Use your own API key (BYOK): Settings → AI Providers → add a key for any of the 145+ providers in the live models.dev catalog (OpenAI, Anthropic, Google, and every OpenAI-compatible endpoint). BYOK models bill against your own provider account and have their own limits.
4. 4Try a local model: Ollama or LM Studio run on your machine with no rate limits.

Atlarix Core (the built-in managed models) is metered as you go. A Pro or Atlarix Reviewer subscription includes an amount of Core usage each month for everyday work, which resets when the subscription renews; beyond that it draws on a prepaid credit balance you top up (from $10), which never expires. On Free, that balance is used from the start.

The allowance is always spent before purchased credit, so the credit figure in Settings is what is left once the month's included usage runs out. If it reaches zero, top it up to keep using Core — or keep working immediately with a BYOK provider or a local model at no cost. You are never locked out — every workspace can fall back to your own keys.

If a specific model keeps failing while others work, it's almost always provider-side — try another model and report the exact error text to support@atlarix.dev so we can investigate.