FileChannel Quickstart

FileChannel writes the conversation to a .md or .txt file, one ## @agent_name section per message. Open the file at any point to see what every agent said. Edit it between runs and the next run continues from your edit. Reference: FileChannel.

Two agents, one file

from swarms import Agent, FileChannel

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

researcher = Agent(
    agent_name="Researcher",
    system_prompt="You gather facts and figures. Answer in a short bullet list.",
    model_name="gpt-5.4",
    max_loops=1,
    channel=channel,
)
writer = Agent(
    agent_name="Writer",
    system_prompt="You turn the facts in the conversation into a short report.",
    model_name="gpt-5.4",
    max_loops=1,
    channel=channel,
)

task = "Report on the EV battery market."
researcher.run(task)
writer.run(task)

print(channel.path.read_text())

The file holds the task once, then each answer:

## @Human

Report on the EV battery market.

## @Researcher

- Global EV battery demand ...

## @Writer

The EV battery market ...

The Writer never receives the Researcher’s answer as an argument. It reads it from the file as a prior turn.

A SequentialWorkflow in a file

Give the channel to the workflow. It posts the task as User, then each agent’s answer in order.

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="Weigh the facts in the conversation and state the main risk.",
            model_name="gpt-5.4",
            max_loops=1,
        ),
        Agent(
            agent_name="Writer",
            system_prompt="Turn the conversation into a three-paragraph brief.",
            model_name="gpt-5.4",
            max_loops=1,
        ),
    ],
    max_loops=1,
    channel=channel,
)
workflow.run("Report on the EV battery market.")

print(channel.path.read_text())

Resume and steer from the file

Every run reads the file first, so a later run, even in a new process, continues from what the file holds. Run the workflow above, then open runs/sequential.md and add a section of your own at the end:

## @Editor

The brief is too long. Keep the risk paragraph and cut the rest to one sentence each.

Then run a follow-up against the same file:

from swarms import Agent, FileChannel, SequentialWorkflow

editor = Agent(
    agent_name="Reviser",
    system_prompt="Apply the editor's notes to the brief in the conversation.",
    model_name="gpt-5.4",
    max_loops=1,
)

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

SequentialWorkflow(agents=[editor], max_loops=1, channel=channel).run(
    "Revise the brief using the editor's note."
)

print(channel.path.read_text())

The Reviser sees the original task, every earlier answer and your note, and its answer is appended under ## @Reviser. You can also delete a weak answer, correct a figure, or remove whole sections before the next run. What is in the file is what the agents see.

AgentRearrange with a parallel step

from swarms import Agent, AgentRearrange, FileChannel

planner = Agent(agent_name="Planner", model_name="gpt-5.4", max_loops=1)
coder = Agent(agent_name="Coder", model_name="gpt-5.4", max_loops=1)
reviewer = Agent(agent_name="Reviewer", model_name="gpt-5.4", max_loops=1)
tester = Agent(agent_name="Tester", model_name="gpt-5.4", max_loops=1)

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

pipeline = AgentRearrange(
    agents=[planner, coder, reviewer, tester],
    flow="Planner -> Coder -> Reviewer, Tester",
    max_loops=1,
    channel=channel,
)
pipeline.run("Write a Python function that validates email addresses.")

print(channel.path.read_text())

The file records User, then Planner, then Coder, then the Reviewer’s and Tester’s answers from the parallel step.

Start over

channel.clear()  # Deletes the file

Tips

Use one file per conversation. Two unrelated runs on the same file see each other’s messages.
Keep ## @ at the start of a line for headers only. A message body that contains such a line is split into two messages when read back.
Check the file into git to keep a reviewable history of what your agents said.

Next steps

Channels with every structure
SqliteChannel, when several processes write at once

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

FileChannel Quickstart

Two agents, one file

A SequentialWorkflow in a file

Resume and steer from the file

AgentRearrange with a parallel step

Start over

Tips

Next steps

​Two agents, one file

​A SequentialWorkflow in a file

​Resume and steer from the file

​AgentRearrange with a parallel step

​Start over

​Tips

​Next steps

Two agents, one file

A SequentialWorkflow in a file

Resume and steer from the file

AgentRearrange with a parallel step

Start over

Tips

Next steps