Skip to main content
davidg426798
davidg426798
April 15, 2026
Question

AI Agent: INVALID_TOOL_RESPONSE

  • April 15, 2026
  • 5 replies
  • 0 views

Sometimes, when I test an agent, it processes for a while, then after a few minutes throws this error:

System Error

This appears to be a system issue. Try again later or submit a support case if the problem continues.

Source: Run Agent - Agent: DG_testAgent - Run ID: 52
Details: LLM execution failed: INVALID_TOOL_RESPONSE - Required schema field not provided in data tree: textValue
Any suggestions on what could be going on? Sometimes the agent succeeds, but the frequency of getting this error makes me concerned about using this in production.
26.3 environment hosted on Appian Cloud.

5 replies

shubhama926776
shubhama926776
April 16, 2026

Most likely a bad/missing output field, the agent is expecting textValue -  always provide a default string fallback.

davidg426798
davidg426798Author
April 16, 2026

Not entirely sure which part you're suggesting for the fix. The agent's output is a record type, and I've given it explicit instructions about how to populate that. Are you suggesting to add a secondary output of type Text, with instructions along the lines of "If you run into any problems, just put a cry for help here"? Or are you instead suggesting that the problem's with one or more of the tools that I've given to the agent? In my current test, those tools are simply record types, which doesn't give me the opportunity for a "string fallback".

shubhama926776
shubhama926776
April 17, 2026

I'm just trying to guess here, as this is new for us too, It’s probably the tool output schema, not the record-type final answer. The missing textValue error means one of the agent’s tool calls is returning data in a shape Appian doesn’t expect, so check the tool/record mapping and required fields first.

S
sam.wahbeh
April 17, 2026

Hi David G.

Do you know if this is failing on the set variable call of the agent run or query record.

This error typically occurs when the LLM hallucinates and doesn't provide the expected schema for the tool call. We are looking at bumping up the number of retries in an upcoming version.

-Sam

davidg426798
davidg426798Author
April 17, 2026

It seemed to be when generating the output record, if I remember correctly--was definitely always after most of the processing had been completed.

Terms & ConditionsAccessibility statement