Skip to content

Quickstart

Quickstart

Run the smallest useful task

from swarms import Agent

agent = Agent(
    agent_name="docs_summarizer",
    system_prompt="You summarize technical notes clearly and briefly.",
    model_name="gpt-4o-mini",
    max_loops=1,
)

result = agent.run("Write three practical checks before adopting a new agent framework.")
print(result)

The first run proves the basic path: the tool starts, reaches the model, and returns an inspectable result. Do not begin with production data, remote entry points, or high-risk tools.

Signal	Meaning
Command or script runs	Environment and package installation are probably correct.
Model call succeeds	Provider configuration is usable.
Output is bounded	The loop or session did not run away.
State can be found	You can recover or debug later.

Make the first run low-risk

Run one Agent with one prompt and one output. Do not add a workflow, memory, tools, or routing until the single script returns a result you can inspect.

python first_agent.py

If this fails	Look here first
Command or import fails	Installation and active environment.
Provider auth fails	Key source, model name, and provider account.
Output is too broad	Prompt, loop limit, and output instruction.
State is hard to find	Session, logs, and working directory.

References

Quickstart: https://docs.swarms.world/quickstart.md
Agent: https://docs.swarms.world/api/agent.md