Fix `UnboundLocalError` and make model configurable

This commit addresses two issues in `software/source/server/i.py`:
1. Fix the `UnboundLocalError` caused by importing the `os` module at both the top level and within the `configure_interpreter` function.
2. Make the `interpreter.llm.model` configurable via the `MODEL` environment variable or `--model` command line argument. Defaults to gpt-4. This allows users to run O1 locally.

Fixes #116

README.md

@@ -44,10 +44,28 @@ cd 01/software # CD into the source directory
 ```shell
 brew install portaudio ffmpeg cmake # Install Mac OSX dependencies
 poetry install # Install Python dependencies
-export OPENAI_API_KEY=sk... # OR run `poetry run 01 --local` to run everything locally
+export OPENAI_API_KEY=sk... 
 poetry run 01 # Runs the 01 Light simulator (hold your spacebar, speak, release)
 ```
 
+### Running locally
+To run locally, you can use command line arguments and environment variables.
+
+Using command line arguments:
+
+```shell
+poetry run 01 --local --model ollama/mixtral:latest
+```
+
+Using command line arguments and environment variables:
+
+```shell
+export MODEL=ollama/mixtral:latest
+poetry run 01 --local
+```
+
+Note, you should replace `ollama/mixtral:latest` with a model installed locally. For supported models, see Open Interpreter's [local providers](https://docs.openinterpreter.com/language-models/local-models/lm-studio) documentation.
+
 <br>
 
 # Hardware

software/source/server/i.py

@@ -295,8 +295,6 @@ def configure_interpreter(interpreter: OpenInterpreter):
     #     if chunk.get("format") != "active_line":
     #         print(chunk.get("content"))
 
-    import os
-
     from platformdirs import user_data_dir
 
 

software/source/server/server.py

@@ -415,6 +415,7 @@ async def main(server_host, server_port, llm_service, model, llm_supports_vision
             globals()[service] = getattr(service_instance, service)
 
         interpreter.llm.completions = llm
+        interpreter.llm.model = os.getenv('MODEL', model)
         
         # Start listening
         asyncio.create_task(listener())