MESSAGE_COERCION_FAILURE

Several modules in LangChain take MessageLike objects in place of formal BaseMessage classes. These include OpenAI style message objects ({ role: "user", content: "Hello world!" }), tuples, and plain strings (which are converted to HumanMessages).

If one of these modules receives a value outside of one of these formats, you will receive an error like the following:

const badlyFormattedMessageObject = {
  role: "foo",
  randomNonContentValue: "bar",
};

await model.invoke([badlyFormattedMessageObject]);

Error: Unable to coerce message from array: only human, AI, system, or tool message coercion is currently supported.

Received: {
  "role": "foo",
  "randomNonContentValue": "bar",
}

Troubleshooting

The following may help resolve this error:

• Ensure that all inputs to chat models are an array of LangChain message classes or a supported message-like.
  • Check that there is no stringification or other unexpected transformation occuring.
• Check the error's stack trace and add log or debugger statements.