Tool calling with runnables #7388
Replies: 1 comment 7 replies
-
|
I found similar issues that might help you resolve the error you're encountering:
To resolve your issue, make sure that for each tool call in your const messageHistory = [
new HumanMessage("message"),
new AIMessage({
content: "",
tool_calls: [
{
name: "toolName",
id: "tool_id",
args: {},
}
],
}),
new ToolMessage({
name: "toolName",
tool_call_id: "tool_id",
content: "string content",
})
];Ensure that each To continue talking to Dosu, mention @dosu. Help Dosu learn! Give it feedback: Great Response | Irrelevant Answer | Incorrect Sources | Too Verbose | Hallucination | Other |
Beta Was this translation helpful? Give feedback.
-
|
@dosu using the code i provided, can you modify that to fix this example you provided |
Beta Was this translation helpful? Give feedback.
-
|
To modify your code to automatically insert a tool message after a tool call, you can adapt the example I provided earlier. Here's how you can integrate it into your existing setup: import { tool } from "@langchain/core/tools";
import { ChatOpenAI } from "@langchain/openai";
import {
ChatPromptTemplate,
MessagesPlaceholder,
} from "@langchain/core/prompts";
import { RunnableWithMessageHistory } from "@langchain/core/runnables";
import { UpstashRedisChatMessageHistory } from "@langchain/community/stores/message/upstash_redis";
import { AIMessage, HumanMessage, ToolMessage } from "@langchain/core/messages";
const llm = new ChatOpenAI({
model: "gpt-4o-mini",
});
const companyListings = `...`; // Your company listings here
const questions = [/* Your questions here */];
const prompt = ChatPromptTemplate.fromMessages([
[
"system",
`You are a professional and friendly Cold-Calling Customer Service Agent...`,
],
new MessagesPlaceholder("history"),
["human", "{input}"],
]);
const analysisTool = tool(
async ({ chatHistory }: { chatHistory: string }) => {
console.log("Questions: \n", questions, "\nAnswers: \n", chatHistory);
return "Analysis complete";
},
{
name: "qulification_analysis",
description: "Used this tool once all the questions have been answered by the user",
}
);
const llmWithTools = llm.bindTools([analysisTool]);
const chain = prompt.pipe(llmWithTools);
const chainWithHistory = new RunnableWithMessageHistory({
runnable: chain,
getMessageHistory: (_sessionId) =>
new UpstashRedisChatMessageHistory({
sessionId: `${_sessionId}-Chat`,
sessionTTL: 10,
config: {
url: process.env.REDIS_URL,
token: process.env.REDIS_TOKEN,
},
}),
inputMessagesKey: "input",
historyMessagesKey: "history",
});
const responses = [
"hi",
"yes i am free to chat",
"yes you can proceed with your questions",
"my budget is $300,000",
"I am looking for a villa",
"I am looking for a villa in Jebel Ali",
"I am looking to move in the next 1 year",
"I need a pool",
"I would like to know more about the company",
];
for (const response of responses) {
console.log(`User input: ${response}\n`);
const responseStream = await chainWithHistory.stream(
{
input: response,
name: "Stacy",
company: "Tech Unicorn",
questions: questions.join("\n"),
companyListings: companyListings,
},
{
configurable: {
sessionId: "123",
},
}
);
const messages = [];
process.stdout.write("AI Response: ");
for await (const chunk of responseStream) {
if (chunk?.tool_calls && chunk?.tool_calls?.length > 0) {
for (const toolCall of chunk.tool_calls) {
const toolMessage = new ToolMessage({
content: "Tool executed successfully.",
tool_call_id: toolCall.id,
status: "success",
});
messages.push(toolMessage);
}
} else {
process.stdout.write(String(chunk.content));
}
}
console.log("\n");
console.log("Tool Messages: ", messages);
}In this code, after processing each response, it checks for any tool calls in the AI's response. For each tool call, it creates a To continue talking to Dosu, mention @dosu. Help Dosu learn! Give it feedback: Great Response | Irrelevant Answer | Incorrect Sources | Too Verbose | Hallucination | Other |
Beta Was this translation helpful? Give feedback.
-
|
@dosu the code you provided does nothing with the messages variable it only stores the tool messages, I still get the following error when I run the code you have provided. error: 400 An assistant message with 'tool_calls' must be followed by tool messages responding to each 'tool_call_id'. The following tool_call_ids did not have response messages: call_7CHFQMAVVdWZ4gpwLvaJsPPm Troubleshooting URL: https://js.langchain.com/docs/troubleshooting/errors/INVALID_TOOL_RESULTS/ Bun v1.1.33 (Windows x64) |
Beta Was this translation helpful? Give feedback.
-
|
To resolve the "INVALID_TOOL_RESULTS" error, you need to ensure that each tool call is followed by a corresponding tool message in the chat history. Here's how you can modify your code to handle this:
Here's an updated version of your code: import { tool } from "@langchain/core/tools";
import { ChatOpenAI } from "@langchain/openai";
import {
ChatPromptTemplate,
MessagesPlaceholder,
} from "@langchain/core/prompts";
import { RunnableWithMessageHistory } from "@langchain/core/runnables";
import { UpstashRedisChatMessageHistory } from "@langchain/community/stores/message/upstash_redis";
import { AIMessage, HumanMessage, ToolMessage } from "@langchain/core/messages";
const llm = new ChatOpenAI({
model: "gpt-4o-mini",
});
const companyListings = `...`; // Your company listings here
const questions = [/* Your questions here */];
const prompt = ChatPromptTemplate.fromMessages([
[
"system",
`You are a professional and friendly Cold-Calling Customer Service Agent...`,
],
new MessagesPlaceholder("history"),
["human", "{input}"],
]);
const analysisTool = tool(
async ({ chatHistory }: { chatHistory: string }) => {
console.log("Questions: \n", questions, "\nAnswers: \n", chatHistory);
return "Analysis complete";
},
{
name: "qulification_analysis",
description: "Used this tool once all the questions have been answered by the user",
}
);
const llmWithTools = llm.bindTools([analysisTool]);
const chain = prompt.pipe(llmWithTools);
const chainWithHistory = new RunnableWithMessageHistory({
runnable: chain,
getMessageHistory: (_sessionId) =>
new UpstashRedisChatMessageHistory({
sessionId: `${_sessionId}-Chat`,
sessionTTL: 10,
config: {
url: process.env.REDIS_URL,
token: process.env.REDIS_TOKEN,
},
}),
inputMessagesKey: "input",
historyMessagesKey: "history",
});
const responses = [
"hi",
"yes i am free to chat",
"yes you can proceed with your questions",
"my budget is $300,000",
"I am looking for a villa",
"I am looking for a villa in Jebel Ali",
"I am looking to move in the next 1 year",
"I need a pool",
"I would like to know more about the company",
];
for (const response of responses) {
console.log(`User input: ${response}\n`);
const responseStream = await chainWithHistory.stream(
{
input: response,
name: "Stacy",
company: "Tech Unicorn",
questions: questions.join("\n"),
companyListings: companyListings,
},
{
configurable: {
sessionId: "123",
},
}
);
const messages = [];
process.stdout.write("AI Response: ");
for await (const chunk of responseStream) {
if (chunk?.tool_calls && chunk?.tool_calls?.length > 0) {
for (const toolCall of chunk.tool_calls) {
const toolMessage = await analysisTool.invoke({ chatHistory: response });
messages.push(new ToolMessage({
content: toolMessage,
tool_call_id: toolCall.id,
}));
}
} else {
process.stdout.write(String(chunk.content));
}
}
console.log("\n");
console.log("Tool Messages: ", messages);
// Append AI message and tool messages to the chat history
chainWithHistory.getMessageHistory("123").add(new AIMessage({
content: response,
tool_calls: messages.map(msg => ({
name: "qulification_analysis",
id: msg.tool_call_id,
args: {},
})),
}));
messages.forEach(msg => chainWithHistory.getMessageHistory("123").add(msg));
}This code ensures that each tool call is followed by a tool message, which is then added to the chat history. This should resolve the error by maintaining the correct sequence of messages [1]. To continue talking to Dosu, mention @dosu. Help Dosu learn! Give it feedback: Great Response | Irrelevant Answer | Incorrect Sources | Too Verbose | Hallucination | Other |
Beta Was this translation helpful? Give feedback.
-
|
To resolve the "INVALID_TOOL_RESULTS" error, you need to ensure that each tool call in the AI's response is followed by a corresponding
Here's an updated version of your code: import { tool } from "@langchain/core/tools";
import { ChatOpenAI } from "@langchain/openai";
import {
ChatPromptTemplate,
MessagesPlaceholder,
} from "@langchain/core/prompts";
import { RunnableWithMessageHistory } from "@langchain/core/runnables";
import { UpstashRedisChatMessageHistory } from "@langchain/community/stores/message/upstash_redis";
import { AIMessage, HumanMessage, ToolMessage } from "@langchain/core/messages";
const llm = new ChatOpenAI({
model: "gpt-4o-mini",
});
const companyListings = `...`; // Your company listings here
const questions = [/* Your questions here */];
const prompt = ChatPromptTemplate.fromMessages([
[
"system",
`You are a professional and friendly Cold-Calling Customer Service Agent...`,
],
new MessagesPlaceholder("history"),
["human", "{input}"],
]);
const analysisTool = tool(
async ({ chatHistory }: { chatHistory: string }) => {
console.log("Questions: \n", questions, "\nAnswers: \n", chatHistory);
return "Analysis complete";
},
{
name: "qulification_analysis",
description: "Used this tool once all the questions have been answered by the user",
}
);
const llmWithTools = llm.bindTools([analysisTool]);
const chain = prompt.pipe(llmWithTools);
const chainWithHistory = new RunnableWithMessageHistory({
runnable: chain,
getMessageHistory: (_sessionId) =>
new UpstashRedisChatMessageHistory({
sessionId: `${_sessionId}-Chat`,
sessionTTL: 10,
config: {
url: process.env.REDIS_URL,
token: process.env.REDIS_TOKEN,
},
}),
inputMessagesKey: "input",
historyMessagesKey: "history",
});
const responses = [
"hi",
"yes i am free to chat",
"yes you can proceed with your questions",
"my budget is $300,000",
"I am looking for a villa",
"I am looking for a villa in Jebel Ali",
"I am looking to move in the next 1 year",
"I need a pool",
"I would like to know more about the company",
];
for (const response of responses) {
console.log(`User input: ${response}\n`);
const responseStream = await chainWithHistory.stream(
{
input: response,
name: "Stacy",
company: "Tech Unicorn",
questions: questions.join("\n"),
companyListings: companyListings,
},
{
configurable: {
sessionId: "123",
},
}
);
const messages = [];
process.stdout.write("AI Response: ");
for await (const chunk of responseStream) {
if (chunk?.tool_calls && chunk?.tool_calls?.length > 0) {
for (const toolCall of chunk.tool_calls) {
// Invoke the tool and create a ToolMessage
const toolMessage = await analysisTool.invoke({
chatHistory: "Your chat history here", // Provide the necessary chat history
});
messages.push(new ToolMessage({
content: toolMessage,
tool_call_id: toolCall.id,
}));
}
} else {
process.stdout.write(String(chunk.content));
}
}
console.log("\n");
console.log("Tool Messages: ", messages);
}This code ensures that each tool call is followed by a To continue talking to Dosu, mention @dosu. Help Dosu learn! Give it feedback: Great Response | Irrelevant Answer | Incorrect Sources | Too Verbose | Hallucination | Other |
Beta Was this translation helpful? Give feedback.
-
Checked other resources
Commit to Help
Example Code
Description
I am trying to build a Q and A bot and once all the questions have been answered I want to send it to the analyser tool to qualify that user, but every time the tool gets used by the LLM i get the following error and I'm not sure how to fix it.
1 | export default "unknown";
^
error: 400 An assistant message with 'tool_calls' must be followed by tool messages responding to each 'tool_call_id'. The following tool_call_ids did not have response messages: call_SvBE9iyZ69QERnYFsoyRC14h
Troubleshooting URL: https://js.langchain.com/docs/troubleshooting/errors/INVALID_TOOL_RESULTS/
Bun v1.1.33 (Windows x64)
System Info
"@langchain/community": "^0.3.12",
"@langchain/core": "^0.3.18",
"langchain": "^0.3.6",
"@langchain/openai": "^0.3.13",
Beta Was this translation helpful? Give feedback.
All reactions