SQL Agent Node Fails to Parse LLM Output Due to Invalid Tool Call

Based on your description and the log output from the SQL Agent node, the issue appears to be that the LLM output isn’t being parsed correctly. Here’s a summary of the problem and some potential next steps:

What’s Happening

• LLM Output Parsing Failure:
  The SQL Agent returns the error “Could not parse LLM output: I don’t know.” This is typically caused by the output not matching the expected schema or format that the agent parser can work with.

• Invalid Tool Call:
  The log shows the agent attempting an action call:

  Action: [[“list-tables-sql”]]
  Observation: [[“list-tables-sql”]] is not a valid tool, try another one.
  

  This indicates that the agent is trying to call a tool with the name “list-tables-sql”, which is not recognized or supported by the node’s current configuration.

Possible Causes

• Misconfiguration of the Agent Prompt:
  The SQL Agent node relies on a carefully set prompt to generate valid queries and tool calls. An incorrect or malformed prompt might cause the LLM to generate unexpected output.

• Mismatch in Tool Definitions:
  The agent might be mistakenly using a tool name it shouldn’t. Since the available tools for SQL interactions are predefined, invoking an invalid one like “list-tables-sql” leads to the output parsing failure.

• Credential or Database Setup Issues:
  Although you’re using Postgres, note that the documentation highlights certain credential tunneling options that aren’t supported for Postgres and MySQL in this agent. However, the log error points more directly to an output parsing/tool naming issue.

Troubleshooting Steps

1. Review Your Prompt Configuration:
   Ensure that the Prompt parameter in the node is set correctly. If you’re using an expression or dynamic input, double-check that it resolves to valid text (i.e. it does not return null).

2. Validate Tool Configuration:
   Confirm that you are using the correct tools. The SQL Agent expects to work with a specific set of tools and tool calls. If you inadvertently use a tool name (such as “list-tables-sql”) that isn’t part of the allowed set, the agent will fail to parse the output.

3. Check Node Options and Parameters:
   Verify other node parameters such as Ignored Tables, Sample Rows, and Prefix/Suffix Prompt to ensure nothing is interfering with the expected LLM output format.

4. Refer to the Documentation:
   The error message links to a troubleshooting guide (OUTPUT_PARSING_FAILURE | 🦜️🔗 Langchain). Reviewing this resource might offer insights or examples similar to your situation.

Summary of Your Setup (for reference)

• n8n version: 1.78.1
• Database: Postgres
• Running via: n8n Cloud
• Operating System: Linux

In conclusion, the primary problem is that the agent is generating an invalid tool call, which causes the output to fail parsing. Adjusting your prompt or verifying the set of allowed tools should be the first steps to take.

Feel free to provide more details if you need further assistance!