[Roadmap] AIOS Roadmap Q3 2024

[Tata0703]

This document outlines the long-term features in the AIOS roadmap for Q3 2024. Feel free to discuss any of the following topics, and add any other topics you'd like to talk about in this issue.

AIOS Documentation (@throwie72)

• manual pages and documentation
• OpenBSD style as an example: https://man.openbsd.org/style

Agent interaction UI (@BRama10 @mannanshukla)

• [ ] Build a UI demo https://github.com/agiresearch/AIOS/issues/133, users can @ a particular agent to indicate that the request needs to be executed by that agent running on AIOS, just like @ somebody on twitter.
  • add multiple commands (following linux commands)
  • add command parser for agent-related operation
  • add output formatter
  • dynamic agent loader
  • request user clarification by agent

Tree/Graph agent workflow support

Support the design of different workflows with different data structures.

• add tree-structured workflow
• add graph-structured (without cycle) workflow
• add graph-structured (with cycle) workflow

Natural-language agent workflow support (@BRama10)

Support non-expert agent developers to use natural language to design and describe their agent workflows.

• reference paper: https://arxiv.org/abs/2405.06907
• integrate natural language workflow programming syntax
• integrate llm interpreter to interpret and execute natural language programmed agents

Automated agent workflow (@itsthemoon)

An extension of Natural-language agent workflow, llm automatically create agent workflow

• add workflow that can be generated by LLM and can be updated

Batchfy agent requests

• batchfy multiple agent requests sent to open-sourced LLM
• batchfy multiple agent requests sent to close-sourced LLM

Support multi-level queues

• add multiple queues for LLM requests
• separate tool calling request and LLM request into different queues

Multi-core AIOS and LLM router

• [ ] Multi-LLM-Core with LLM router which can direct agent request to the most cost-efficient/time-efficient LLM https://github.com/agiresearch/AIOS/issues/129

[om-raheja]

Easy way to generate agents off of yaml file/fine tuned LLM, and then you can just download agents from somewhere

To provide isolation between agents and editing the AIOS source code.

[evison]

Support for customized LLMs fine-tuned by agent developers.

[evison]

Defining a set of API standards and implementing a microservices architecture, you can create a scalable, flexible, and maintainable system for agent development. break down the system into modular components that would optimize time and resources by a lot -- Giovanni