AgentOoba is an autonomous AI agent extension for Oobabooga’s web ui.

AgentOoba is a very new project for implementing an autonomous agent in Oobabooga’s web UI. It does so by making detailed requests to the underlying large language model.

The agent takes a “divide and conquer” approach to the task: if it cannot find a suitable way to accomplish the goal immediately, it tries to break the task into subtasks and evaluates each subtask in a breadth-first recursive manner.

AgentOoba was designed with a small context model in mind. Its hinting system is designed to try to break down general hints into smaller sub-hints, giving the model only the context each hint absolutely needs. This allows for smaller context sizes at the cost of longer execution times.

  • NOTE: This project is still in its infancy. Now the agent is able to use the tool and use the built-in capabilities of the model to accomplish the task, but it is not good at it. It needs more context.
  • The latest update includes changes to how task streams are handled. Previously, the agent would try to use the tool to complete the task as soon as it encountered the tool; now, it waits for the subtask to complete.


start up

1. Use options --extensions AgentOoba Start Oobabooga, you can do this by editing the startup script; python (additional arguments) line should be changed to python --extensions AgentOoba (additional arguments) .

You can also start it normally and go to the extensions tab to enable AgentOoba, but you have to do this every time you start it.

2. Load Model – The agent is designed to be flexible to the model type, but the labor and helper prefixes must be set according to the model type in the prompt section of the UI. These are currently set for the Wizard series of models.

3. Go to the main text generation page and scroll down to see the UI.

#AgentOoba #Homepage #Documentation #Downloads #Oobabooga #Autonomous #Agent #Extension #News Fast Delivery

Leave a Comment

Your email address will not be published. Required fields are marked *