Multi-Agent Systems

Elsai provides several patterns for orchestrating multiple agents:

Pattern	Use when
Graph	You know the execution order ahead of time (deterministic pipeline)
Swarm	Agents should self-organise and decide who does what (autonomous)
Workflow	Complex sequential stages with conditional logic or step-level error recovery
Agent as Tool	Simplest form — one agent calls another like any other tool
A2A	Agents live in separate services and communicate over the network

Why multi-agent?

• Parallelism — run independent tasks concurrently
• Specialisation — each agent has a focused role, tools, and system prompt
• Context management — distribute work across smaller context windows
• Scalability — compose complex workflows from simple building blocks

Graph (deterministic)

Agents are nodes in a directed graph. Edges define dependencies and data flow.

from elsai import Agent
from elsai.multiagent import Graph

researcher = Agent(name="researcher", description="Searches and summarises information")
writer = Agent(name="writer", description="Writes polished articles based on research")
editor = Agent(name="editor", description="Reviews and improves written content")

graph = Graph()
graph.add_node(researcher)
graph.add_node(writer)
graph.add_node(editor)

# writer receives researcher's output; editor receives writer's output
graph.add_edge(researcher, writer)
graph.add_edge(writer, editor)

result = graph("Write an article about the future of AI")
print(result)

Swarm (autonomous)

Agents collaborate as a team. A coordinator decides which specialist handles each part.

from elsai import Agent
from elsai.multiagent import Swarm

coordinator = Agent(
    name="coordinator",
    description="Routes tasks to the right specialist"
)
coder = Agent(name="coder", description="Writes Python code")
reviewer = Agent(name="reviewer", description="Reviews code for bugs and style")

swarm = Swarm(agents=[coordinator, coder, reviewer])
result = swarm("Write and review a Python function that sorts a list of dicts by a key")
print(result)

Agent as tool

The simplest form of multi-agent: wrap an agent as a tool for another agent.

from elsai import Agent

researcher = Agent(
    name="researcher",
    description="Searches for information on any topic and returns a summary"
)

writer = Agent(
    name="writer",
    tools=[researcher.as_tool()]
)

result = writer("Write a blog post about quantum computing")

Related

• Graph — deterministic pipeline with declared dependencies
• Swarm — autonomous agent collaboration
• Workflow — imperative sequential/parallel pipelines
• Agent as Tool — the simplest multi-agent pattern
• Agent-to-Agent (A2A) — cross-service agent communication