using local ollama models returns a json structure instead of executing them

[jwcloud365]

Describe the bug

Today I installed openinterpreter and want to use it locally.

But when I'm using a local ollama model, I get some json structure as an output:

Loading qwen2.5-coder:32b...

Model loaded.

Open Interpreter will require approval before running code.

Use interpreter -y to bypass this.

Press CTRL-C to exit.

list all files in my Images directory

{"name": "execute", "arguments":{"language": "shell", "code": "dir %USERPROFILE%\Pictures"}}

browse to cnn.com

  {"name": "execute", "arguments":{"language": "python", "code":
  "computer.browser.setup(headless=False)\ncomputer.browser.go_to_url('https://www.cnn.com')"}}

But when i use an openai model, I get this output, as expected:

list all files in my Images directory

ls ~/Images

Would you like to run this code? (y/n)

If you need further actions or details on any specific file, please let me know!

browse to cnn.com

  computer.browser.go_to_url('https://www.cnn.com')


  Would you like to run this code? (y/n)

Reproduce

interpreter

list all files in my Images directory

{"name": "execute", "arguments":{"language": "shell", "code": "dir %USERPROFILE%\Pictures"}}

browse to cnn.com

  {"name": "execute", "arguments":{"language": "python", "code":
  "computer.browser.setup(headless=False)\ncomputer.browser.go_to_url('https://www.cnn.com')"}}

Expected behavior

  computer.browser.go_to_url('https://www.cnn.com')


  Would you like to run this code? (y/n)

Screenshots

I would expect

instead of

{"name": "execute", "arguments":{"language": "python", "code":
"computer.browser.setup(headless=False)\ncomputer.browser.go_to_url('https://www.cnn.com')"}}

Open Interpreter version

Version: 0.4.3

Python version

Python 3.11.11

Operating System name and version

W11

Additional context

No response

[thomahn3]

Same here using phi, llama3.2 and mistral on macos 15.2.

[nainglinwai1]

For those problems, I found the same error after first try. I am using the python 3.10. I can run with llama3.2:latest model and open-interptreter local tutorial. It's working now.

[66Ton99]

The same, but it works OK with qwen

[nainglinwai1]

I found the solution to make it works.
First you need to export OLLAMA_MODEL with this
export OLLAMA_MODEL=llama3.2:latest
export OLLAMA_API_BASE=http://localhost:11434/v1

Run these before you run the code.

[thomahn3]

I found the solution to make it works. First you need to export OLLAMA_MODEL with this export OLLAMA_MODEL=llama3.2:latest export OLLAMA_API_BASE=http://localhost:11434/v1

Run these before you run the code.

I ran this code before running interpreter and now I get the following error: 16:39:34 - LiteLLM:ERROR: utils.py:1873 - Model not found or error in checking vision support. You passed model=llama3.2:latest, custom_llm_provider=ollama. Error: OllamaError: Error getting model info for llama3.2:latest. Set Ollama API Base via 'OLLAMA_API_BASE' environment variable. Error: Client error '404 Not Found' for url 'http://localhost:11434/v1/api/show/api/show'