6. Troubleshooting and Support

16. Error handling and troubleshooting

Error or symptom	Likely cause	Recommended action
401 with LiteLLM_VerificationTokenTable or token_not_found_in_db	A new Logicc key is being sent to the former LiteLLM endpoint.	Use https://api.logicc.cloud/v1 and remove the former api.logicc.io base URL.
401 Authorization failed or virtual_key_not_found	The key is missing, invalid, revoked, or sent in the wrong header.	Verify the complete key and Authorization header. Request a replacement if the key previously worked.
GET /v1/models works but a required model is missing	The key is valid, but the model or provider route is not enabled for it.	Send the required model IDs to Logicc support. The models endpoint is authoritative for each key.
429 or RESOURCE_EXHAUSTED	A request, token, provider, or account quota has been reached.	Reduce concurrency, retry with exponential backoff, and inspect the quota endpoint before contacting support.
messages.*.tool_use.input must be a valid dictionary	A tool call contains a string, list, or malformed JSON instead of an object.	Ensure every tool input is a valid JSON object, including empty inputs represented as {}.
Function tools with reasoning_effort are not supported	The selected model does not support that parameter combination on chat completions.	Set reasoning_effort to none for this request or use a compatible model and endpoint.
A third-party client fails while cURL works	The client may override the base URL, headers, model ID, or request schema.	Compare its outgoing request with the cURL example and confirm custom OpenAI-compatible endpoints are supported.

For production workloads, configure explicit connection and response timeouts, retry only transient failures, cap retry attempts, and preserve idempotency where a repeated request could trigger an external action.

17. Production checklist

Store the key in a secret manager or protected environment variable.

Verify the base URL and authentication headers in every environment.

Discover enabled model IDs through GET /v1/models during setup.

Set explicit timeouts and bounded retries for 429 and transient 5xx responses.

Validate tool inputs as JSON objects before sending requests.

Log request time, endpoint, model ID, status code, and request ID without logging secrets or sensitive content.

Monitor the quota endpoint and technical rate limits.

Test model-specific features such as streaming, tools, files, images, and structured output before release.

Define fallback behavior for unavailable models and provider interruptions.

18. Support

Include the following information for fast technical support:

Company and contact person

Environment and client or SDK version

Base URL and complete endpoint

Model identifier

Approximate timestamp and timezone

HTTP status code and complete error body

Minimal reproducible request with all secrets removed