How to enable automatic file creation from LLM responses (JSON actions not being executed)

[BraZucco]

I'm trying to use Continue.dev with a local model (Ollama) to create and edit files automatically based on LLM responses.

However, when the model returns structured JSON indicating a file creation action, Continue does not execute it. Instead, it treats the response as plain text JSOIN

{
  "name": "create_new_file",
  "arguments": {
    "filepath": "./teste.py",
    "contents": ""
  }
}

[pandego]

What you are seeing is expected if the model is only emitting JSON text.

Continue will not execute arbitrary JSON blobs from chat output as file actions. For file creation/editing you need the agent/tool path, and with Ollama that usually means both:

• a model that actually supports tool calling/function calling
• capabilities: [tool_use] enabled in your Continue model config when auto-detection is not enough

The Continue docs call out that Agent mode with Ollama depends on tool support, and the Ollama config page shows tool_use as the capability to enable for models that support it.

So the rough rule is:

• plain JSON in the assistant response -> treated as text
• tool call emitted through the supported agent/tool interface -> Continue can act on it

If you want, post your config.yaml model block plus the exact Ollama model id you are using. I can help sanity-check whether the blocker is missing tool support vs config.