ollama - 💡(How to fix) Fix When using a model that is capable of tools and having it hallucinate a non-existent tool, the server throws a 500?

What is the issue?

I am testing various Ollama models for a personal assistant project that uses tool calling. While testing Ministral-3:3b, the model occasionally hallucinates tool calls for tools that are not present in the provided tool list.

This behavior from the model itself is expected and not the issue.

The problem is that when the model outputs a nonexistent tool call, the server responds with an HTTP 500 error instead of returning the model output normally or producing a handled validation/tool-not-found response.

Since model output should be treated as untrusted input, hallucinated or invalid tool calls should ideally be handled gracefully without causing an internal server error.

Expected behavior:

• Return the model response as-is, letting the model call the non-existing tool and letting the rest of the code take care of it

Actual behavior:

• Server returns HTTP 500 when the model calls a nonexistent tool.

This makes it difficult for client applications to implement their own fallback or validation handling. Please note this is for the Ollama python library

Relevant log output

OS

macOS 26.5

GPU

Apple M4, 8 core

CPU

Apple M4, 8 core

Ollama version

0.20.7