[FEATURE] Graph Agent

[ylwu-amzn]

Is your feature request related to a problem?
Currently we have three types of agent (doc):

• flow agent
• Conversational flow agent
• Conversational agent

These existing agent types have limitations in supporting complex logic. Specifically, they lack the ability to:

• Implement conditional workflows
• Execute tools in parallel
• Handle branching and merging of execution paths
• Manage dependencies between tasks
• Human in the loop

This limitation restricts the creation of more sophisticated and efficient workflows within OpenSearch.

What solution would you like?
We propose implementing a new type of agent that supports a graph-style workflow. This Graph-Based Workflow Agent would:

1. Allow users to define workflows as directed acyclic graphs (DAGs)
2. Support conditional branching based on task outputs or predefined conditions
3. Enable parallel execution of independent tasks
4. Provide mechanisms for synchronization and merging of parallel branches
5. Allow for dynamic task generation and insertion during runtime
6. Offer better visibility and control over the workflow execution process
7. Human-in-the-Loop (HITL): Manual intervention for approvals, exception handling, and workflow tuning.
8. Decision-making, dynamic workflow modifications, and prompt generation.
9. Multi-Agent Collaboration: Role-based agents with distinct memory spaces collaborate for modular workflows.

This new agent type would significantly enhance OpenSearch's capability to handle complex, multi-step processes and decision-making workflows.

What alternatives have you considered?
Enhance current flow type.

[pyek-bot]

I would like to work on this, can you please assign this to me?

[ramda1234786]

@ylwu-amzn , this solution is planned for which release? Thanks

[ramda1234786]

@ylwu-amzn and @pyek-bot can we add following also in the agent

1. Human-in-the-Loop (HITL): Manual intervention for approvals, exception handling, and workflow tuning.
2. Decision-making, dynamic workflow modifications, and prompt generation.
3. Multi-Agent Collaboration: Role-based agents with distinct memory spaces collaborate for modular workflows.

and of course the short term and long term memory management

[ylwu-amzn]

@ylwu-amzn and @pyek-bot can we add following also in the agent

1. Human-in-the-Loop (HITL): Manual intervention for approvals, exception handling, and workflow tuning.
2. Decision-making, dynamic workflow modifications, and prompt generation.
3. Multi-Agent Collaboration: Role-based agents with distinct memory spaces collaborate for modular workflows.

and of course the short term and long term memory management

@ramda1234786 Thanks for your suggestion. Add these to the plan.

[ylwu-amzn]

@ylwu-amzn , this solution is planned for which release? Thanks

Not decided yet. Researching now. Will share a plan once we have effort estimation done.