ai-agents-ollama-json-mode-bug-using-ollama-with-response-format-type-json-object-97da2ccf

THEN Avoid using response_format with Ollama models until the fix is released. Instead, omit response_format and parse the raw JSON from the model response manually. Alternatively, apply a local patch to litellm/llms/ollama/completion/transformation.py to detect JSON-only responses and skip function call extraction.

# Workaround: don't pass response_format
response = completion(model="ollama/phi3:latest", messages=[{"role": "user", "content": "What is the capital of France?"}])
# Then parse response.choices[0].message.content as JSON