List of BaseMessages with consecutive runs of message types merged into single messages, or a RunnableLambda which returns a list of BaseMessages If two messages being merged both have string contents, the merged content is a concatenation of the two strings with a new-line separator. If at least one of the messages has a list of content blocks, the merged content is a list of content blocks.
import { mergeMessageRuns, AIMessage, HumanMessage, SystemMessage, ToolCall } from "@langchain/core/messages";
const messages = [
new SystemMessage("you're a good assistant."),
new HumanMessage({ content: "what's your favorite color", id: "foo" }),
new HumanMessage({ content: "wait your favorite food", id: "bar" }),
new AIMessage({
content: "my favorite colo",
tool_calls: [{ name: "blah_tool", args: { x: 2 }, id: "123" }],
id: "baz",
}),
new AIMessage({
content: [{ type: "text", text: "my favorite dish is lasagna" }],
tool_calls: [{ name: "blah_tool", args: { x: -10 }, id: "456" }],
id: "blur",
}),
];
mergeMessageRuns(messages);
The above example would return:
[
new SystemMessage("you're a good assistant."),
new HumanMessage({
content: "what's your favorite colorwait your favorite food",
id: "foo",
}),
new AIMessage({
content: [
{ type: "text", text: "my favorite colo" },
{ type: "text", text: "my favorite dish is lasagna" },
],
tool_calls: [
{ name: "blah_tool", args: { x: 2 }, id: "123" },
{ name: "blah_tool", args: { x: -10 }, id: "456" },
],
id: "baz",
}),
]
Generated using TypeDoc
Merge consecutive Messages of the same type.
NOTE: ToolMessages are not merged, as each has a distinct tool call id that can't be merged.