47163e4b-1af6-4552-bb79-7eb4151cde40 fe8618f1-bfb3-44f5-b02f-00b3ed0f06a0 d981ce4f-84d0-45d4-a64a-78346774f0b5 d9e0270d-9e46-4fe8-91d2-d0b13e0ce798 @Shubham Bajaj can you help us debug what when wrong with these calls? They all had the same error and we see this issue failry often.

Hi there! 👋 I see you're encountering the

call.in-progress.error-vapifault-custom-llm-llm-failed

error across multiple calls. While I can't debug those specific call IDs without more information, I can definitely help explain what this error means and provide some general debugging steps that might help resolve the issue. ## What this error means The error

call.in-progress.error-vapifault-custom-llm-llm-failed

indicates a failure related to your custom LLM integration with Vapi. According to [information from Discord](https://discord.com/channels/1211482211119796234/1211483291191083018/threads/1335181348285054996), this error may be related to: - LLM failures - Issues with custom LLM usage - Token limits being exceeded - Errors from services like 11Labs ## Possible causes There are several potential causes for this error: 1. **LLM Failures**: Service outages, network issues, or problems with specific requests being made to the LLM. 2. **Custom LLM Configuration Issues**: Your LLM key might not be correctly configured or might have exceeded usage limits set by the LLM provider. 3. **Token Limits**: The error can be triggered if token limits are being exceeded or if there's incorrect token calculation. 4. **Incompatible Response Format**: If your custom LLM response doesn't match the expected OpenAI-compatible format, particularly if it's missing the "choices" array property or if the property is not an array. 5. **Tool Use and Result Formatting**: Incorrectly formatted or sequenced tool use and tool result blocks. 6. **Model-Specific Issues**: If you're using 11labs flash_model, [one Discord thread](https://discord.com/channels/1211482211119796234/1211483291191083018/threads/1335277468805758986) suggests changing to the turbo model might help. ## Debugging steps Here are some steps you can take to debug this issue: 1. **Review configurations and logs**: - Check your server logs for more detailed error messages

🔵 21:38:35:644 Completion Request Timeout (#1, provider: custom-llm, model: gpt-4o, region: undefined, credential: true) 🔵 21:38:35:646 Couldn't Complete Model Sequence. Request Timed Out And Sequence Empty (provider: custom-llm, model: gpt-4o, region: undefined, credential: true) 🔵 21:38:15:644 CustomLLMRequest Messages: [ { "role": "assistant", "content": "Hola, soy mía. ¿En qué puedo ayudarte hoy?" }, { "role": "user", "content": "Quería hacer una cita." }, { "role": "assistant", "content": "Claro, ¿estás buscando ver a algún proveedor en específico?" }, {"role": "user", "content": "No."}, { "role": "assistant", "content": "Perfecto. ¿Eres un paciente nuevo o ya has visitado la clínica antes?" }, {"role": "user", "content": "No, no soy nuevo."}, { "role": "assistant", "content": "Gracias. ¿Tienes algún día o horario en mente que te funcione mejor para la cita?" }, { "role": "user", "content": "Pues, si habría hoy o el día que sea." }, { "role": "assistant", "content": "Hoy no hay disponibilidad, pero el miércoles dieciséis de abril doctor Patrick Pack tiene espacio a las tres quince pm. Si prefieres algo más temprano, jueves diecisiete de abril hay un horario disponible a las once quince am. ¿Te interesa alguno de estos?" }, {"role": "user", "content": "De temprano."}, { "role": "assistant", "content": "Perfecto, antes de confirmar la cita, ¿me podrías proporcionar tu nombre completo, número de teléfono y fecha de nacimiento?" }, { "role": "user", "content": "Sí, es siete catorce tres seis tres veinte seis nueve. Fecha de nacimiento es enero diecisiete dos mil doce." }, { "role": "assistant", "content": "Gracias. ¿Me podrías confirmar tu nombre completo, por favor?" }, {"role": "user", "content": "Ross."}, { "role": "assistant", "content": "Gracias, Ross. Para confirmar, ¿estás agendando una cita con el doctor Patrick Bach el jueves diecisiete de abril a las once quince AM. ¿Es correcto?" }, {"role": "user", "content": "Correcto."}, { "role": "assistant", "content": "Dame un momento para agendar tu cita. Necesito cita. Necesito tu apellido para completar la cita. ¿Me lo podrías proporcionar, por favor?" }, {"role": "user", "content": "Ok. It's wings."} ]

I see

Yes, we create timers before requests. If the time expires without a response, we notify the user of the delayed message or request.

Documentation examples are not up to date on this topic. I got it working on a cloudflare worker. Only the ```openai-advanced```file was succesful when copy pasting. https://github.com/VapiAI/example-server-serverless-cloudflare/blob/main/src/api/custom-llm/openai-advanced.ts

This ticket has been marked as solved, and as a result, you will no longer receive any further responses. Kindly create a new support ticket to escalate your inquiry.