Did anyone tryout agent swarm from openai, it just came out this friday. I feel it is still very nascent and does not look like as robust as langgraph. I did not see ways to create different interaction patterns and also did not see any shared persistent state.
1 Like
From the GitHub it’s not meant to be used in production and instead is
intended to explore ergonomic interfaces for multi-agent systems.
Swarm is an educational resource for developers curious to learn about multi-agent orchestration. Swarm runs (almost) entirely on the client and, much like the Chat Completions API, does not store state between calls.
Maybe they were interested in it and dropped it later? Who knows.
1 Like
from swarm import Swarm, Agent
client = Swarm()
def transfer_to_agent_b():
return agent_bagent_a = Agent(
name=“Agent A”,
instructions=“You are a helpful agent.”,
functions=[transfer_to_agent_b],
)agent_b = Agent(
name=“Agent B”,
instructions=“Only speak in Haikus.”,
)response = client.run(
agent=agent_a,
messages=[{“role”: “user”, “content”: “I want to talk to agent B.”}],
)print(response.messages[-1][“content”])
Results:
They talk only when they want?
Is this related to this one: https://openai.com/search/?q=SWARM
?
Is this the repository: GitHub - openai/swarm: Educational framework exploring ergonomic, lightweight multi-agent orchestration. Managed by OpenAI Solution team.?
It’s not an official OpenAI product nor will it be supported by OpenAI. It “was” a repo by an OpenAI dev who was building some stuff to investigate use cases. Think of it as a personal project repo. Use it for inspiration by all means, but don’t use it as the basis for your next big project 
2 Likes
It is an OpenAI project because it was released on the openai account on GitHub.
OpenAI should assume everything they do is not only about good repositories for pumping stocks.
Well, I see OpenAI does not play it safe and gets inspired by other projects (probably like they did with training data…). Ilya was right… 
I will bring more for inspiration in my project too: all people should have the **OPTION TO** use their Private AIs with Private Data and should **CHOSE** what Open Source Models and other Closed Source Models they want to use.
1 Like
I recommend this video on Swarm by Sam Witteveen:
"you tube"watch?v=npAljHBeKPc
Sorry, I’m new here and at “trust level 0”, thus cannot add links yet - haha, all good.
1 Like
Opeai Swarm works with AzureOpenAI? Thoughts?
1 Like
I’ve been working with Swarm in Node.js to manage and orchestrate dozens of specialized agents for our project, hoping to streamline agent management and task delegation. The idea was to use a master agent to route tasks to the appropriate agents and improve latency by running the orchestration at the OpenAI endpoint.
However, when attempting to transfer tasks between agents, we receive the following LLM response:
“I’m sorry for the confusion, but as an artificial intelligence, I do not have the capacity to transfer you to a specific agent since we don’t possess individual identities like human agents. However, I’m here to assist you with any queries or issues you might have. Please, let me know how I can help you.”
It appears that agent-to-agent task transfer is not supported, which is critical for our workflow. After digging into it further, I learned that Swarm is not an official OpenAI product and is instead a personal project by an OpenAI developer. This might explain why certain core features, such as dynamic task handoff between agents, are incomplete or missing.
While Swarm still holds great potential for orchestrating complex, multi-agent tasks, I’m wondering if anyone has suggestions for workarounds or whether others have faced similar challenges when trying to use Swarm in production environments? It’s clear that this repo may not be fully production-ready, but I’m curious if there are any paths forward for those of us trying to implement multi-agent orchestration at scale.
2 Likes
Any reason why you are not considering langgraph
1 Like
Exploring it now! How’s your experience with specialized workflows?
Can you please share what you mean by specialized workflows
1 Like
Part 4: Specialized Workflows
https://langchain-ai.github.io/langgraph/tutorials/customer-support/customer-support/#car-rental-tools```
1 Like
@Revhooman Take a look at GitHub - ax-llm/ax: The unofficial DSPy framework. Build LLM powered Agents and "Agentic workflows" based on the Stanford DSP paper.
Its an agentic framework that handles a lot of workflows, and has a fast router that handles routing requests to the correct agent built in called fast-llm-router
1 Like