Integrating OpenRouter into Live Helper Chat with Responses API and Tool Calls support

OpenRouter is a unified API gateway that gives you access to hundreds of AI models — including OpenAI, Anthropic Claude, Google Gemini, Meta Llama, and many others — through a single OpenAI-compatible endpoint.

Before you begin, you will need:

• An API Key from https://openrouter.ai/keys
• A configured REST API in Live Helper Chat
• A configured Bot in LHC

tip

OpenRouter uses an OpenAI-compatible Responses API (/api/v1/responses). The LHC configuration files below are nearly identical to the ChatGPT Responses API integration — the only differences are the host (https://openrouter.ai) and the model name.

Streaming vs Non-Streaming

Choose the variant that fits your infrastructure:

Variant	Requires
Non-Streaming	PHP only
Streaming	PHP + NodeJS + PHP-Resque extensions

Flow with Tool Call Support

The examples below include two sample tool-call functions (password_reminder and support_price) to demonstrate how function calling works end-to-end.

REST API

• Import one of the REST API configurations:
  • Non-Streaming REST API
  • Streaming REST API
• Set the Bearer token value to your OpenRouter API key (YOUR_API_KEY).
• Change the model field in the request body to the model you want to use, for example openai/gpt-4o, anthropic/claude-3.5-sonnet, google/gemini-2.0-flash etc. See the full list at https://openrouter.ai/models.
• Adjust the system message content to match your use case.
• Optionally replace YOUR_VECTOR_STORAGE_ID with your OpenAI vector store ID if you want file search. Remove the file_search tool entry if you don't need it.
• After testing you may want to uncheck Log all request and their responses as system messages.
• If responses take a long time, increase Maximum execution time.

Request body overview

The request body sent to OpenRouter looks like this (simplified):

{
    "model": "openai/gpt-4o",
    "stream": false,
    "parallel_tool_calls": false,
    "input": [
      {
        "role": "system",
        "content": "You are a helpful assistant..."
      }
    ],
    "tools": [
        {
            "type": "file_search",
            "vector_store_ids": ["<YOUR_VECTOR_STORAGE_ID>"]
        },
        {
            "type": "function",
            "name": "password_reminder",
            "description": "Sends a password reset link to the provided email address.",
            "parameters": {
                "type": "object",
                "properties": {
                    "email": {
                        "type": "string",
                        "description": "The email address associated with the account."
                    }
                },
                "required": ["email"]
            }
        },
        {
            "type": "function",
            "name": "support_price",
            "description": "Returns paid support price",
            "parameters": {
                "type": "object",
                "properties": {},
                "required": []
            }
        }
    ]
}

Set "stream": true when using the streaming REST API import.

Bot

• Import one of the bot configurations:
  • Non-Streaming Bot
  • Streaming Bot
• During the import wizard, select the REST API you just imported.

Calling a trigger based on a defined function

This is how you would wire up the transfer_operator scenario (the same pattern applies to any tool call).

1. Define the function in the REST API request body:

{
  "type": "function",
  "name": "transfer_operator",
  "description": "Transfers to a human agent if the user explicitly asks for one or the bot cannot resolve the issue.",
  "parameters": {
    "type": "object",
    "properties": {},
    "required": []
  }
}

You can also add a hint to your system prompt:

Always call transfer_operator function if (and only if) you are explicitly asked for a human (or agent) OR if you cannot help the user with a specific task.

2. Add an event to the relevant trigger in your bot:
   • Type: Custom text matching
   • Should include any of these words: transfer_operator

Limiting the knowledge base to uploaded documents (Vector Storage)

Here are example system instructions for a documentation bot:

You are a helpful Live Helper Chat Bot. You answer questions based on file search. If you don't know the answer, respond with "I can only help with Live Helper Chat related questions." Provide the most relevant answer to the visitor's question, not exceeding 100 words. Include a link for more information about your answer.

Choosing a model

OpenRouter supports a large number of models. Change the model field in the REST API request body to switch models without any other changes. Some popular options:

Provider	Model identifier
OpenAI	openai/gpt-4o
OpenAI	openai/gpt-4.1
Anthropic	anthropic/claude-3.5-sonnet
Anthropic	anthropic/claude-3.7-sonnet
Google	google/gemini-2.0-flash
Meta	meta-llama/llama-3.3-70b-instruct
DeepSeek	deepseek/deepseek-chat

See the full model list at https://openrouter.ai/models.

Debugging

For debug purposes, edit the REST API in the back office and check Log all request and their responses as system messages. This will display the raw API request and response as system messages in the chat, which makes it easy to diagnose unexpected outputs or failed tool calls.