Variable convertResponsesMessageToAIMessage`Const`

convertResponsesMessageToAIMessage: BaseDynamicToolInput<
    ResponsesCreateInvoke
    | ResponsesParseInvoke,
    BaseDynamicToolInput,
> = ...

Converts an OpenAI Responses API response to a LangChain AIMessage.

This converter processes the output from OpenAI's Responses API (both create and parse methods) and transforms it into a LangChain AIMessage object with all relevant metadata, tool calls, and content.

Param: response

The response object from OpenAI's Responses API. Can be either:

ResponsesCreateInvoke: Result from responses.create()
ResponsesParseInvoke: Result from responses.parse()

Returns

An AIMessage containing:

id: The message ID from the response output
content: Array of message content blocks (text, images, etc.)
tool_calls: Array of successfully parsed tool calls
invalid_tool_calls: Array of tool calls that failed to parse
usage_metadata: Token usage information converted to LangChain format
additional_kwargs: Extra data including:
- refusal: Refusal text if the model refused to respond
- reasoning: Reasoning output for reasoning models
- tool_outputs: Results from built-in tools (web search, file search, etc.)
- parsed: Parsed structured output when using json_schema format
- Function call ID mappings for tracking
response_metadata: Metadata about the response including model, timestamps, status, etc.

Throws

Error if the response contains an error object. The error message and code are extracted from the response.error field.

Example

const response = await client.responses.create({
  model: "gpt-4",
  input: [{ type: "message", content: "Hello" }]
});
const message = convertResponsesMessageToAIMessage(response);
console.log(message.content); // Message content
console.log(message.tool_calls); // Any tool calls made

Remarks

The converter handles multiple output item types:

message: Text and structured content from the model
function_call: Tool/function calls that need to be executed
reasoning: Reasoning traces from reasoning models (o1, o3, etc.)
custom_tool_call: Custom tool invocations
Built-in tool outputs: web_search, file_search, code_interpreter, etc.

Tool calls are parsed and validated. Invalid tool calls (malformed JSON, etc.) are captured in the invalid_tool_calls array rather than throwing errors.