I have a very complex workflow developed in langgraph and I want to attach VAPI voice capabilities. Every request the VAPI receives from the user has to be redirected to the langgraph agent. Could you pleae help me out how to configure this. Thanks

To connect VAPI to your LangGraph agent, configure VAPI to use your LangGraph endpoint as a Custom LLM:

bash
curl -X PATCH https://api.vapi.ai/assistant/YOUR_ASSISTANT_ID \
     -H "Authorization: Bearer YOUR_PRIVATE_KEY" \
     -H "Content-Type: application/json" \
     -d '{
  "model": {
    "provider": "custom-llm",
    "model": "gpt-4o",
    "url": "https://YOUR_LANGGRAPH_ENDPOINT/chat/completions"
  }
}'

If your LangGraph agent is running locally, use ngrok to create a public URL. Sources: - [Custom LLM Tool Calling Integration](https://docs.vapi.ai/documentation/advanced-concepts/customization/custom-llm/custom-llm-tool-calling-integration) - [Using Your Server](https://docs.vapi.ai/customization/custom-llm/using-your-server) - [Server SDKs](https://docs.vapi.ai/server-sdks)

how to setup in dashboard?

According to [Using Your Server](https://docs.vapi.ai/customization/custom-llm/using-your-server), you can set up LangGraph in the VAPI dashboard by: 1. Log in to your VAPI account at [dashboard.vapi.ai](https://dashboard.vapi.ai/) 2. Navigate to the "Model" section 3. Choose the "Custom LLM" option 4. Enter your LangGraph endpoint URL in the endpoint field (use ngrok URL if your LangGraph agent runs locally) 5. Test the connection by sending a test message Is your LangGraph endpoint already accessible via a public URL, or do you need guidance on setting up ngrok for a locally running LangGraph agent? Source: - [Using Your Server](https://docs.vapi.ai/customization/custom-llm/using-your-server)

I've added my custom URL ( https://b456-2605-ad80-30-241e-f9a5-117c-9942-7f2c.ngrok-free.app), but still its adding /chat/completions in the last which makes the system not working

Create a custom Vapi tool that sends user voice input via POST to your LangGraph API endpoint, which processes it and returns a JSON response for Vapi to vocalize.

Yes, you need to have OpenAI structured response for CustomLLM and it must be

chat/completions

It also needs to be streaming response. https://github.com/VapiAI/example-custom-llm

Hey Sai, checking if this was resolved for you.