ollama-json-fixer-python

Force local Ollama models to output strict, valid JSON. Zero dependencies beyond requests.

Problem

Local LLMs (Llama 3, Mistral, Qwen, Codellama) often wrap JSON in conversational text or markdown fences, breaking json.loads(). This causes JSONDecodeError in production pipelines.

Solution

• Auto-strips markdown fences
• Regex extraction of JSON blocks
• Auto-retry with strict system prompt
• Works with ANY Ollama model
• Single file - just copy and use

Quick Start

pip install requests
python ollama_json_fixer.py