transformeroptimus / superagi Goto Github PK

I've been playing around with other language models such as GPT4all, Vicuna 13B, Alpaca and the results have been promising so far. Is there any way for me to test SuperAGI with other open source models? That would be great!

I should have an option to define LTM or Long term memory by picking any of the vector dbs - Pinecone, Weaviate, Chroma etc

Resource Manager for agent

An agent can have its own task queue to pick and execute tasks. For agent clusters, there can be a shared task queue accessible to multiple agents.

Need log management in SuperAGI. There can be different types of logs like agent logs, system logs etc.

There should be multi-modal agents that can process images, audio files, etc so that we can provide more context along with the prompt.

Need a way to stream agent response. Eg. Knowledge based LLM responses
Can be structured similar to ChatGPT

An action console for agent feedback.

The agent could be running or waiting for user feedback. There should be a dedicated way to give feedback to the agent.

Hey Guys

I just managed to install SuperAGI and it's working as it should.

I was thinking of testing a a case where I can write personalized job application emails with my Gmail. Is there any way to connect SuperAGI with gmail?

It would be nice to see if agents are able to create sub-agents to delegate tasks assigned.

Should be part of Input Resource module. Similar to #67

Need a Budget Manager manage to manage costs like api usage

Has a tool been made to parse PDFs into SuperAGI? I'd like to give SuperAGI the ability to access a PDF file.

The potential benefits are listed below:

1. As syntactic norms are followed or that certain words or phrases are avoided, restructuring prompts to LLMs using LMQL should greatly reduce the JSON errors that popup (will be even more relevant as the capabilities are added)
2. LMQL maintains or improves the accuracy of various downstream tasks while substantially reducing computation or cost in pay-to-use APIs, resulting in cost savings >10%

Introductory research paper: https://arxiv.org/pdf/2212.06094.pdf
Website: https://lmql.ai/
Docs: https://docs.lmql.ai/
Example: https://github.com/rumpfmax/Multi-GPT/tree/master/multigpt/lmql_utils

There should be a way to create a cluster of agents which can pick tasks from task queue and run in parallel.

Would be cool to have Github as a tool

You can connect your github repo, it should be able to read the code and then make pull requests to make edits to code

One of the challenge with existing libraries for autonomous agents is that they get in loop because they don't know if the task they are trying to accomplish has been completed. So explicit option to define exit condition for agent can be a good feature to include.

Exit condition can be user defined or LLM defined.

Same primitive can be extended to include time based ( TAT ) or steps based exit.

We need to create a marketplace for community tools with guidelines on how to contribute. The user can see the number of tools installs, upvote, downvote, add comments etc.

What if we can have Weviate and Mivus as vector databases along with Pinecone?

SuperAGI command line. The args and params can be similar to REST apis

Agents should have the ability to perform Google Search

Need a way for this agent to support multiple languages

Need a way to deploy agents to cloud

We need a protocol for agent to communicate with other agents.

Hey

Is there a feature to manage the budget? I would like to know how many of my credits are being used in a single run.

Hi guys

I was just wondering if we can make the agents conversational - as in each agent communicates with each other in order to get the assign tasks completed ?

Need a way to create and manage different types of agents with tasks, without tasks, chatbot etc.

If Agent goes into a loop, it should detect it and find ways to come out of it and proceed to the next step.

Need to finalise the agent creation, execution, project, org etc. architecture. And add the diagram in docs as well.

There should be a way for users to see list of all tools built for SuperAGI and install them in their project with one click

There should be a way for agents to fine tune their trajectory based on their behaviours in simulated environments without using the actual tools

I'm trying to connect SuperAGI to Twitter so that the agent can scout for the latest trends in AI, summarize that and create automated tweets with the summary. Is there any way to connect Twitter?

There should be a way to provision, spawn and run agents through a GUI

GUI should include :

1. Ability to provision an agent
2. Ability to run an agent and see the agent feed, have an action console to see all the actions / notifications, stats of run
3. Create tools - it can include code packaging and bundling as well from gui

Hey guys !

Can we also have a way to let agents have long-term memory? like using Pinecone to store the context in the database?

Does this support multi modal agents ?