Note: The example automatically downloads the `gpt4all` model as it is under a permissive license. The GPT4All model does not seem to be enough to run AutoGPT. WizardLM-7b-uncensored seems to perform better (with `f16: true`).
## Without docker
Run AutoGPT with `OPENAI_API_BASE` pointing to the LocalAI endpoint. If you run it locally for instance:
```
OPENAI_API_BASE=http://localhost:8080 python ...
```
Note: you need a model named `gpt-3.5-turbo` and `text-embedding-ada-002`. You can preload those in LocalAI at start by setting in the env: