Code interpreters agent but not with openai

[Nevermetyou65]

Hi, I want to build code interpreter agent like in this medium
https://medium.com/@venugopal.adep/building-an-ai-data-analysis-assistant-with-llamaindex-and-openai-c0e371a432d6

but instead of using openai, I want to use qwen3 instead. Can someone guide me the way please?

[onestardao]

Hey — good question, and the Medium article is a nice walkthrough.

To replicate the code interpreter agent pattern using llama-index but swap out OpenAI with Qwen3, here’s a minimal viable direction:

1. Replace the OpenAI LLM wrapper
   LlamaIndex supports custom LLMs via llms.OpenAI–style abstractions.
   You’ll want to use the custom LLM wrapper to point to your local/inference endpoint for Qwen3 (or via HuggingFace pipeline, etc.)

2. Ensure tool-calling / code interpreter capabilities
   If you’re using something like AgentRunner or OpenAIAgent, you’ll need to confirm that your Qwen3 backend has function-calling or tool-usage features implemented (many Qwen3 variants don’t natively support this yet, unless wrapped).
   You might need to fallback to explicit tool routing via ToolExecutor.

3. Runtime setup (sandboxing optional)
   If the agent is executing Python code, be cautious: the Medium article assumes a controlled code interpreter (like OpenAI Code Interpreter). With Qwen3, you’ll likely need to define a safe execution sandbox manually (e.g. using exec, subprocess, or tools like jupyter_client).

Optional: if you're debugging LLM-agent interaction logic across tools and backends, I've also compiled a public AI Reasoning Problem Map — covers lots of weird edge cases when mixing LLMs + toolchains.

Let me know if you'd like a minimal working Qwen3 agent starter — happy to help!

— onestardao (WFGY)