Channels with Every Structure

Nine multi-agent structures and SwarmRouter take a channel argument. The structure posts the task once, then each member’s contribution under that member’s name, so the channel becomes a readable record of who said what. Every example on this page uses a FileChannel so you can open the result. Replace it with any other channel and the code stays the same:

from swarms import FileChannel, QueueChannel, SqliteChannel, HttpChannel

channel = FileChannel("runs/team.md")
# channel = QueueChannel()
# channel = SqliteChannel("channels.db", name="team")
# channel = HttpChannel("http://127.0.0.1:8765/channels/team")

Give the channel to the structure only. A structure with a channel detaches the channels of its member agents, with a warning, so each message is posted once.

At a glance

Structure	What lands in the channel
SequentialWorkflow	`User`, then each agent in order
ConcurrentWorkflow	`User`, then every agent in agent order once the step completes
AgentRearrange	`User`, then each agent as its step of the flow completes
GraphWorkflow	Each node’s output as its layer completes; no task row
MixtureOfAgents	`User`, each worker per layer as `Name (layer i/n)`, then the aggregator
HierarchicalSwarm	`User`, the director’s plan and orders, each worker’s report
GroupChat	`User`, then each message that wins a turn
MajorityVoting	`user`, each voter, then `Consensus-Agent`
PlannerWorkerSwarm	`User`, the plan, each worker’s result, the `CycleJudge` verdict
SwarmRouter	Whatever the structure it builds posts

SequentialWorkflow

from swarms import Agent, FileChannel, SequentialWorkflow

channel = FileChannel("runs/sequential.md")

workflow = SequentialWorkflow(
    agents=[
        Agent(agent_name="Researcher", system_prompt="Gather the key facts as a short bullet list.", model_name="gpt-5.4", max_loops=1),
        Agent(agent_name="Analyst", system_prompt="State the main risk in the facts above.", model_name="gpt-5.4", max_loops=1),
        Agent(agent_name="Writer", system_prompt="Turn the conversation into a short brief.", model_name="gpt-5.4", max_loops=1),
    ],
    max_loops=1,
    channel=channel,
)
workflow.run("Report on the EV battery market.")

## @User
Report on the EV battery market.

## @Researcher
...

## @Analyst
...

## @Writer
...

Each agent also reads what the channel already held before the run, so a second run on the same file continues the conversation.

ConcurrentWorkflow

from swarms import Agent, ConcurrentWorkflow, FileChannel

channel = FileChannel("runs/concurrent.md")

workflow = ConcurrentWorkflow(
    agents=[
        Agent(
            agent_name=f"{angle.title().replace(' ', '-')}-Analyst",
            system_prompt=f"Analyse the task from the {angle} angle only.",
            model_name="gpt-5.4",
            max_loops=1,
        )
        for angle in ["technology", "regulation", "supply chain"]
    ],
    channel=channel,
)
workflow.run("What will shape the EV battery market over the next five years?")

Each agent receives only the task. The channel records the three answers in agent order, not in the order they finished.

AgentRearrange

from swarms import Agent, AgentRearrange, FileChannel

channel = FileChannel("runs/rearrange.md")

pipeline = AgentRearrange(
    agents=[
        Agent(agent_name="Planner", model_name="gpt-5.4", max_loops=1),
        Agent(agent_name="Coder", model_name="gpt-5.4", max_loops=1),
        Agent(agent_name="Reviewer", model_name="gpt-5.4", max_loops=1),
        Agent(agent_name="Tester", model_name="gpt-5.4", max_loops=1),
    ],
    flow="Planner -> Coder -> Reviewer, Tester",
    max_loops=1,
    channel=channel,
)
pipeline.run("Write a Python function that validates email addresses.")

The channel records User, Planner, Coder, then Reviewer and Tester from the parallel step.

GraphWorkflow

from swarms import Agent, FileChannel, GraphWorkflow

channel = FileChannel("runs/graph.md")

analyst = Agent(agent_name="Analyst", model_name="gpt-5.4", max_loops=1)
writer = Agent(agent_name="Writer", model_name="gpt-5.4", max_loops=1)
researcher = Agent(agent_name="Researcher", model_name="gpt-5.4", max_loops=1)
editor = Agent(agent_name="Editor", model_name="gpt-5.4", max_loops=1)

wf = GraphWorkflow(channel=channel)
for agent in [analyst, writer, researcher, editor]:
    wf.add_node(agent)  # The node id is the agent's name

wf.add_edge("Analyst", "Writer")
wf.add_edge("Analyst", "Researcher")
wf.add_edge("Writer", "Editor")
wf.add_edge("Researcher", "Editor")
wf.set_entry_points(["Analyst"])
wf.set_end_points(["Editor"])

wf.run(task="Produce a short market report on AI chips.")

GraphWorkflow posts each node’s output, layer by layer: Analyst, then Writer and Researcher, then Editor. It does not post the task.

MixtureOfAgents

from swarms import Agent, FileChannel, MixtureOfAgents

channel = FileChannel("runs/moa.md")

moa = MixtureOfAgents(
    agents=[
        Agent(agent_name="Security", system_prompt="Review from a security angle.", model_name="gpt-5.4", max_loops=1),
        Agent(agent_name="Reliability", system_prompt="Review from a reliability angle.", model_name="gpt-5.4", max_loops=1),
        Agent(agent_name="Cost", system_prompt="Review from a cost angle.", model_name="gpt-5.4", max_loops=1),
    ],
    aggregator_agent=Agent(
        agent_name="Lead-Reviewer",
        system_prompt="Combine the reviews into one prioritised list.",
        model_name="gpt-5.4",
        max_loops=1,
    ),
    layers=2,
    channel=channel,
)
print(moa.run("Review our plan to move the billing database to a managed Postgres service."))

## @User
...
## @Security (layer 1/2)
...
## @Reliability (layer 1/2)
...
## @Cost (layer 1/2)
...
## @Security (layer 2/2)
...
## @Lead-Reviewer
...

HierarchicalSwarm

from swarms import Agent, FileChannel, HierarchicalSwarm

channel = FileChannel("runs/hierarchical.md")

swarm = HierarchicalSwarm(
    agents=[
        Agent(agent_name="DataWorker", agent_description="Finds and summarises data.", model_name="gpt-5.4", max_loops=1),
        Agent(agent_name="WritingWorker", agent_description="Writes clear prose from notes.", model_name="gpt-5.4", max_loops=1),
    ],
    director_model_name="gpt-5.4",
    max_loops=1,
    channel=channel,
)
swarm.run("Produce a short competitive analysis of the AI chip market.")

The channel records the task, the director’s plan and orders, and each worker’s report under its name. Internal System notes stay out of the channel.

GroupChat

from swarms import Agent, FileChannel, GroupChat

channel = FileChannel("runs/groupchat.md")


def speaker(name, stance):
    return Agent(
        agent_name=name,
        system_prompt=f"You argue for {stance}. Keep each message short.",
        model_name="gpt-5.4",
        max_loops=1,
    )


chat = GroupChat(
    agents=[
        speaker("Optimist", "the benefits"),
        speaker("Pessimist", "the risks"),
        speaker("Realist", "a balanced view"),
    ],
    max_loops=6,
    threshold=0.5,
    channel=channel,
)
chat.run("Should hospitals adopt AI for triage?")

Only messages that win a turn are posted. The private bids each agent makes are not. GroupChat adds the respond tool the agents need to bid.

MajorityVoting

from swarms import Agent, FileChannel, MajorityVoting

channel = FileChannel("runs/vote.md")

vote = MajorityVoting(
    agents=[Agent(agent_name=f"Voter-{i}", model_name="gpt-5.4", max_loops=1) for i in range(1, 6)],
    channel=channel,
)
vote.run("Is Python or Rust better for a high-throughput web server? Answer in one word, then one sentence.")

The channel records the task under user, each voter’s answer, then the Consensus-Agent’s verdict.

PlannerWorkerSwarm

from swarms import Agent, FileChannel
from swarms.structs.planner_worker_swarm import PlannerWorkerSwarm

channel = FileChannel("runs/planner_worker.md")

swarm = PlannerWorkerSwarm(
    agents=[
        Agent(agent_name="Market-Worker", agent_description="Market sizing and competitors.", model_name="gpt-5.4", max_loops=1),
        Agent(agent_name="Pricing-Worker", agent_description="Pricing and packaging.", model_name="gpt-5.4", max_loops=1),
    ],
    planner_model_name="gpt-5.4",
    max_loops=1,
    channel=channel,
)
swarm.run("Draft a go-to-market plan for a B2B analytics product.")

The channel records the task, the planner’s plan, each worker’s result under its name, and the CycleJudge verdict.

SwarmRouter

from swarms import Agent, FileChannel, SwarmRouter

channel = FileChannel("runs/router.md")

router = SwarmRouter(
    agents=[
        Agent(agent_name="Analyst", model_name="gpt-5.4", max_loops=1),
        Agent(agent_name="Writer", model_name="gpt-5.4", max_loops=1),
        Agent(agent_name="Reviewer", model_name="gpt-5.4", max_loops=1),
    ],
    swarm_type="SequentialWorkflow",
    max_loops=1,
    channel=channel,
)
router.run("Write a short blog post about transformer architectures.")

SwarmRouter forwards the channel for these swarm_type values:

Forwards the channel	Ignores it
`SequentialWorkflow`, `ConcurrentWorkflow`, `AgentRearrange`, `MixtureOfAgents`, `HierarchicalSwarm`, `GroupChat`, `MajorityVoting`, `PlannerWorkerSwarm`	`HeavySwarm`, `CouncilAsAJudge`, `MultiAgentRouter`, `LLMCouncil`, `DebateWithJudge`, `RoundRobin`

Next steps

CommsChannel reference: the run lifecycle, the message format and writing your own channel
QueueChannel: watch any of these structures live

Index

Basic Examples

Model Providers

Single Agent

Tree of Thoughts

Multi-Agent Examples

Comms Channels

Auto Agent Builder

Scheduled Agents

Applications

Research

Technical Analysis

Use Cases

Finance

Voice Agents

Integrations

MCP Tool Integrations

Deployment

CLI

Channels with Every Structure

At a glance

SequentialWorkflow

ConcurrentWorkflow

AgentRearrange

GraphWorkflow

MixtureOfAgents

HierarchicalSwarm

GroupChat

MajorityVoting

PlannerWorkerSwarm

SwarmRouter

Next steps

​At a glance

​SequentialWorkflow

​ConcurrentWorkflow

​AgentRearrange

​GraphWorkflow

​MixtureOfAgents

​HierarchicalSwarm

​GroupChat

​MajorityVoting

​PlannerWorkerSwarm

​SwarmRouter

​Next steps

At a glance

SequentialWorkflow

ConcurrentWorkflow

AgentRearrange

GraphWorkflow

MixtureOfAgents

HierarchicalSwarm

GroupChat

MajorityVoting

PlannerWorkerSwarm

SwarmRouter

Next steps