Troubleshooting 'Plan and Execute Agent' Node Errors in n8n

Below is an overview of the issue, along with some pointers on what might be causing it and how to troubleshoot:

---

Error Messages

You’re seeing one (or both) of these error messages when using the Plan and Execute Agent node:

1. Error:
   Problem in node ‘AI Agent’
   Cannot read properties of undefined (reading ‘text’)

2. Error:
   Problem in node ‘AI Agent’
   Unable to parse JSON response from chat agent.
   Question: Can you provide a step-by-step guide on how to create a new Next.js project using the npx create-next-app command?
   ...
   Troubleshooting URL: OUTPUT_PARSING_FAILURE | 🦜️🔗 Langchain

---

What These Errors Mean

• Undefined ‘text’ Property:
  This error points to a scenario where the code expected a response object (or a field within it) to contain a text property, but instead it received an undefined value. This often indicates that the data coming from the chat agent is not in the expected format.

• JSON Parsing Failure:
  The JSON response from the chat agent couldn’t be parsed. This can happen if the response is malformed, missing expected keys/values, or if there’s a discrepancy in the output format.

---

Possible Causes

1. Mismatched or Missing Node Configuration:

   • The Plan and Execute Agent node might be configured in a way that it expects the prompt to be manually defined (rather than being taken from a previous node).
   • Ensure that the prompt is set correctly and that any referenced fields or expressions are returning valid, non-null values.

2. Output Format Issues:

   • The AI agent (powered by LangChain) is expected to return a JSON blob that includes specific fields (such as text). If the underlying API call or response transformation does not return the data in the anticipated schema, parsing will fail.
   • This can also occur if there’s an issue with the conversion of the raw response to JSON.

3. Tool/Sub-Node Connection Problems:

   • Make sure that all required sub-nodes (for example, your Chat Model node) are connected properly. Some agent nodes require connecting additional sub-nodes to manage aspects like memory or API calls.

---

Troubleshooting Steps

• Verify Prompt Settings:
  Check if the Prompt parameter is set to Define below rather than taking input automatically from the previous node. If expressions are used, ensure they refer to valid data and do not resolve to null.

• Inspect Sub-node Connections:
  Confirm that you have connected all the required sub-nodes such as a Chat Model. If the node isn’t connected correctly, the agent might not receive the proper inputs for generating its response.

• Review the Response Format:
  Using the troubleshooting documentation linked in the error message (OUTPUT_PARSING_FAILURE) can be very useful here. This guide explains potential output parsing issues and how to adjust your API calls or prompts so that the response conforms to the required JSON structure.

• Test with Simple Prompts:
  Try isolating the problem by testing the node with a simple, static prompt. This can help determine if the error originates from dynamic or improperly constructed expressions.

• Check for Updated Node Versions:
  If you’re using the Plan and Execute Agent node along with other agent types, ensure that you’re using the latest version of the node. Occasionally, issues arise from outdated sub-nodes which might have incompatibilities.

---

Following these steps should help you pinpoint whether the problem is with your prompt, the connection of sub-nodes, or the format of the output being generated. Adjust the configuration accordingly and retest your workflow.

Based on the error messages you shared, it appears that the Plan and Execute Agent node (which is powered by the AI Agent node) is not receiving the prompt or valid JSON data from its chat agent. Here are some steps you can take to troubleshoot the issue:

1. Check the Prompt Input

• If you're using the "Take from previous node automatically" option, verify that the incoming data contains the expected property (usually named chatInput or similar).
• Remove any null values or update the previous node’s output so that it contains a valid, non-null prompt.
• Alternatively, switch to the "Define below" option so you can directly set the prompt. This helps ensure that your prompt is correctly formed and not missing.

2. Verify JSON Response Structure

• The error "Cannot read properties of undefined (reading 'text')" suggests that the response from the chat agent doesn't include the expected text field.
• Ensure that the chat model or underlying AI component returns properly formatted JSON where the response includes a text (or similar) property that the node expects.
• If needed, consider adding a function or set node to inspect and reformat the response so that it conforms to the expected schema.

3. Check for Node Version and Sub-Node Connectivity

• Confirm that your workflow is using the latest version of all sub-nodes (e.g., Chat Model, Window Buffer Memory, etc.). If you suspect an outdated sub-node, try removing and re-adding it to ensure you’re using the latest version.
• Ensure that any connectors required (like the Retriever or Chat Model connections) are correctly configured in your workflow.

4. Review the Chat Agent’s Output Parsing

• The error "Unable to parse JSON response from chat agent" often indicates that the response content might not be valid JSON or is missing expected fields.
• Double-check any responses that the chat agent returns. If you see unexpected formatting, use debugging nodes (such as a Set node or a Function node) to log and inspect the JSON output.

Summary

In summary, verify that:

• Your prompt is valid and not null.
• The input field from the previous node is correctly named (or manually defined).
• The response JSON includes the expected text property.
• All connected sub-nodes (Chat Model, etc.) are up to date and properly connected.

Following these steps should help you narrow down the issue and get the Plan and Execute Agent node functioning properly.