[FEATURE] Agent Graph to allow streamed response #608
Description
Problem Statement
I would like Strands to add streaming support to an agent graph so that I can enable an increased user-experience on the receiving end while still getting the power of agent graphs conditions and parallel processing.
Proposed Solution
Extend the graph with an stream_async in the same way as an agent has it today.
e.g. something like
from strands import Agent
from strands.multiagent import GraphBuilder, Swarm
# Create a swarm of research agents
research_agents = [
Agent(name="medical_researcher", system_prompt="You are a medical research specialist..."),
Agent(name="technology_researcher", system_prompt="You are a technology research specialist..."),
Agent(name="economic_researcher", system_prompt="You are an economic research specialist...")
]
research_swarm = Swarm(research_agents)
# Create a single agent node too
analyst = Agent(system_prompt="Analyze the provided research.")
# Create a graph with the swarm as a node
builder = GraphBuilder()
builder.add_node(research_swarm, "research_team")
builder.add_node(analyst, "analysis")
builder.add_edge("research_team", "analysis")
graph = builder.build()
#NOTE the proposed "stream_async"
result = graph.stream_async("Research the impact of AI on healthcare and create a comprehensive report")Use Case
When needing to use the power of an agent graph in a more controlled workflow approach, but when still having an end user on the other side where we want to have a higher degree of user experience.
Example Use case: Take a text (e.g. prompt to research, text to summarise etc) and then e.g.
- Validate text against criteria. Only continue if valid
- Run below parallel
- Extract e.g. actions from a text
- Extract decisions from a text
- Extract summary overview from a text
- Look up (http) links for research in parallel
- Gather results from previous steps in the agent graph
- Validate certain criteria or quality
and then have the final "node" in the graph with the ability to stream back the response to a real user
Alternatives Solutions
No response
Additional Context
No response