AC AGI is a proof of concept of leveraging BabyAGI and human feedback to complete a users tasks. This what built in less than 24hrs for the LabLab.ai x Langchain hackathon
Leveraging the Langchain library, ScriptKit and ChatGPT, I've created a system that sets off on a user provided task and queries the user for input when it gets a question it needs help with.
It has the ability to Google for results as well as scrape a Website for the information.
install https://www.scriptkit.com/ install script here
you'll be prompted to install libs and enter api-tokens follow steps there for api info
- Enter your task
- Wait for the agent to complete the task
- Assign max-iterations for the agent to loop: 0 for infinite (probably not a good idea ¯_(ツ)_/¯)
- Profit
Known issues:
- The agent will sometimes get stuck in a loop and not complete the task
- Human feedback is not always helpful
Upcoming features:
- More tools
- Refined prompts
- Better human feedback system
- Better memory system
Possible thanks to the fine folks at Langchain and all the other giants whose shoulders we stand on.