Uffe is...
- a sub-100 lines LLM agent.
- fitted with long term memory.
- self-conscious in the sense that all the code in uffe.py is injected into the system prompt.
- going to use your terminal. Whatever you have access to, uffe has access to.
- a real doer, he will probably not ask for permission, he just gets things done.
- completely independent. You can run uffe.py on vanilla python.
uffe_demo.mp4
- Clone this repository.
- Ensure you have a
.envcontaining OPENAI_API_KEY. - Optional: Make an alias 'uffe' pointing to 'python /path/to/uffe.py'
Without alias:
$ python <path/to/uffe.py> <whatever you want him to do>
With alias:
$ uffe <whatever you want him to do>
If you don't want to manually confirm every console command you can set environment variable
SAFEMODE to 0. You can do this by either putting it in your .env-file or simply:
$ SAFEMODE=0 uffe <whatever you want him to do>
Goes without saying but although this repository is completely open source and free, OpenAI will charge you money for using their LLM. With safemode turned off, there is no limit of how much tokens uffe will use, it can very well get stuck in an infinite loop and drain your OpenAI account.
Use with caution, this agent can (and probably will) do unexpected things using your computer. It has access to whatever you have access to. Use with caution. We do not take responsibility for any damages caused by using this.
- Improve memory