Hey, how's it going today? I want to show you Forest Swarm. So this is a very interesting swarm architecture that we

basically designed from the ground up to allow you to dynamically send tasks, and then this task will be routed

to the most specialized agent for that specific task. And so basically, it's kind of like a router, where you have a task and then

this task is then routed to an agent that is specialized, given its prompt, to this task. And I'll show you how to get

started with it. Um, but yeah, so this is basically how you use it, and I'm going to go to the cursor really quickly

and show you how to run this. So, um, basically you import from `swarms` import `TreeAgent`, `Tree`, `ForestSwarm`, and

then you build your tree. So there's `Trees`, and then there's a `Forest`. The forest has multiple trees, but for

every tree, you can have multiple agents. Um, let's see here if... this is the diagram. So, um, basically we

have a forest, and the forest is kind of like everything; it's every agent. But then within specific trees, there are

different agents. And so when you give it a task, the task will be routed to an agent that is most specialized for that

task. Um, so it's kind of like an agent... it's kind of like an agent storage,

it's kind of like an agent storage, but it's also like dynamic task routing. Um, and I'll show you how to get started

with this now. Um, so if you copy this example... from `swarms` import `ForestSwarm`, `Tree`, `TreeAgent`.

Here we're building our trees. We have... this is mostly like a

financial swarm, but we could also easily make this specialized so we can repurpose this for medical

and healthcare task so we could give this to Claude and then we could say, um, let's create a medical forest, uh,

specialize all of these agents for different, uh, types of healthcare, uh, diagnosis specialist agents, um, and treatment. Let's build three

trees and make the prompts extensive. And so we're going to have, uh, Claude here generate a new

forest and then, uh, we'll we'll run that and I'll basically show you how it works. So we'll call this medical forest

swarm and, uh, it's going to build the, uh, the agency here. And so we have a primary

care, a primary diagnostician, dietician, we have a lab analyst, imaging specialist, we have a treatment planning agent,

medication manager, intervention specialist, and then we have a follow-up and monitoring tree. So we have a

recovery monitor, we have a prevention specialist, and, uh, we have a patient educator, um, and yeah, that's that's our

tree so far. So we're going to copy this, um, and then we're going to try running this and see, uh, what what happens

precisely. So, as I mentioned before, you give it a task. So here we're saying, "Patient presents with persistent

headaches for two weeks, accompanied by visual disturbances and neck stiffness. A comprehensive evaluation and treatment plan is needed." So we're going to go here,

we're going to run this, and then we're going to see all of the agents collaborating with, uh, collaborating with

each other and working with each other, uh, to basically help route this task, uh, to the nearest agent. And so once once we run this, okay, so we

got to be more specific with the file path. Um, what was this one? Yeah, tree form. Okay, now it should

work. And actually, let me clean those outputs. So this is really useful for when you have a bunch of agents and you don't

Know when to send a task to which agent. Um, essentially, what's going on in the backend is we have a RAG system

that will route your task to the nearest agent, uh, based off of cosine similarity. Um, and so, uh, here it's

running. We're going to basically embed all of the agents and all of their prompts, and then we're going to

essentially, uh, get them to, um, to, uh, execute on the task. So, uh, the agent that was

selected is the primary diagnostician, and so this task that we just gave it, uh, was

routed: Patient presents with persistent headaches for two weeks, accompanied by visual disturbances and neck stiffness. A need for comprehensive evaluation and

treatment plan. So what happened here is that, uh, this task was routed to, uh, this agent, the primary diagnostician.

And so here we have the output: Patient assessment – persistent headache, visual disturbances, neck stiffness. These

symptoms raise concern for potential serious conditions, such as meningitis, subarachnoid hemorrhage, or

intracranial mass lesion. Preliminary diagnosis considerations: The combination of headache, neck stiffness,

and visual disturbances may suggest meningitis, especially if accompanied by fever. Subarachnoid hemorrhage: Sudden

onset of severe headache, often described as a "thunderclap" headache, with neck stiffness and visual

changes could also indicate a subarachnoid hemorrhage. Intracranial mass: Persistent headaches and I'll

essentially give you a couple diagnosis possibilities. Then it also gives you next steps: Immediate

referral to emergency department due to the presence of red flags, particularly the combination of headache, visual disturbances, and neck stiffness.

Evaluation in emerging settings is recommended. A thorough neurological examination could be conducted to assess for any focal

neurological deficits. Um, and that's basically kind of like a, uh, very simple demo, but we can

also repurpose this for financial trading. So we're going to take this; now we're going to

say, uh, now let's build an all-new forest, let's build an all-new forest specialized for different

types of financial trading: mutual funds, index funds, um, what else? ETFs, and energy and AI stocks.

And so now, Cole is going to build a new forest that we can use to basically conduct task

completions upon very specialized queries. So, where it's building a mutual fund analyst, it's building

an index fund specialist, it's building an ETF strategist, and an energy sector analyst, an AI technology analyst.

And then we have a couple of other portfolio strategists, technical analysts, or risk managers. Um, and then we can copy this and

then we can say, uh, fund manager forest. Um, and so now we have our fund manager. And so now, Claude should have

been given its task: to analyze current opportunities in AI sector ETFs, considering market conditions and provide a risk-adjusted portfolio

allocation strategy. Um, so let's see what happens here. Oh, so this is probably not going to run because we need to put the full file

path. And, um, oh yeah, I forgot. Okay, so I forgot to say how to download this. Um, so you could download swarms with

pip install das-capital-u swarms. Um, and that should install. You could also assume, as you download swarms,

it's going to ask you to put in some environment variables. So you could do export workspace directory and the

The workspace directory is essentially where all the agents will log all their outputs.  So here, within

the forest, you can also put a model name. So if you put a model name, then you can use—in this example I'm using GPT-4,

but you could use Claude 2.  However, you would need to put in an Anthropic

API key or OpenAI key. It defaults to OpenAI. So you would have to

do it like this and then put your key there. I'm not going to do that because I already have my key, but

that's the onboarding process. So yeah, so we have this workflow. Cloud build here. We have how

many trees? We have three trees. Okay, and we're going to run this task:  Analyze current

opportunities in AI sector ETFs, considering market conditions, and provide a risk-adjusted portfolio allocation strategy. So we're going to run that,

and then we're going to see the agent complete the task. Now, in the future, what we're probably going to do is

to make this better, to make ForestSwarm better, we're probably going to add the ability to add multiple forests. So if

you wanted to run, like, if you wanted to have, like, an entire company marketing a product—the whole company—

Oh, we're getting an error. What is the error? Oh yeah, okay. So that's that's yeah,

that's Claude. I don't have a Claude API key there, but in the future we're

probably going to allow you to have multiple forests, and then you can combine them together

through, kind of like a, kind of like a mega-forest, or I've got to find a word for

It, but yeah, basically, uh, you could have multiple forests, and then, uh, the agents would, um, like one agent would run and

then the next agent would dynamically find what other agent to contact, uh, based off of the query, um, or we could

find a way to chain them together, um, but that's in the near future. Uh, so the agent that ran is

called the mutual fund analyst. And so the AI sector has been, uh, seen significant growth in interest due to

the advancement of machine learning, uh, data and analytics, and automation. However, the sector is subject to high volatility and rapid technological

changes. So, historical returns: analyze past performance of AI sector ETFs over the time frames—one year, three years—to verify

your returns to understand the growth trajectory. Volatility matters. Um, examine the standard deviation and beta of these ETFs to assess their volatil—uh,

volatility relative to the broader market. Fund manager evaluation: access the track record of fund managers in managing technology or thematic funds,

consistency and strategy, experience, and navigating the market. Side note: expense ratios and fee structures. Compare the expense ratios of different

AI sectors. So this basically, uh, saying like how to, like, analyze, like, how would you, um, you know, like, invest into a

major fund of AI stocks—things and sector allocation. Review the top holdings within each ETF to

ensure diversification across different AI sub-sectors. Fund inflows and outflows: monitor recent inflows or

outflows as indications of investor sentiment and liquidity. Uh, risk-adjusted return: Sharpe ratio. Evaluate the risk-adjusted return of each ETF to

understand how well it compensates investors for the risk taken. Sortino ratio: focus on downside risk by analyzing the Sortino

ratio, which only considers negative volatility. Um, and then recommendations: core allocation—uh, 60% invest in broad

AI sector ETF with a strong track record, low expense ratio, and diversified holdings. This provides stable exposure to the AI sector.

Tactical allocation: Allocate to a specialized AI ETF focusing on high-growth sub-sectors such as AI hardware or cloud-based AI services to capture potential upside.

Defensive allocation: Consider an ETF with established tech companies with significant AI investments for downside protection; for example, Facebook, Amazon, Google, etc.

Risk management: Regularly rebalance your portfolio to maintain the desired allocation.

Now, we'll be more specific.  This is a broad outline; now I'll provide a list of ETFs.

I'll add the names of the best, reliable AI ETFs that align with this strategy, and include where to purchase them.

We're going to run it again and hopefully it will be much more specific. This is a tactical and strategic outline of a plan; hopefully, the next results will be more precise, with actual ETFs to buy, etc.

It's running, processing...there you go.  This is from a mutual fund analyst, etc. So, recommended...

Recommended AI sector ETFs: Global X Robotics & AI ETF (BOTZ). Strong historical returns with moderate volatility (approximately 0.68). Holdings focus on companies involved in robotics and AI technologies.

Purchase locations: Available on major platforms like Vanguard, Fidelity, and Charles Schwab.

Building Forests of Agents: Introducing The ForestSwarm 🤖🌲

Kye Gomez

ForestSwarm 소개 및 기본 개념 설명

ForestSwarm이란 동적으로 작업을 전달하고 가장 특화된 에이전트에 라우팅하는 군집 아키텍처

작업은 특정 작업에 특화된 에이전트로 라우팅됨

라우터와 유사하게 작업을 특화된 에이전트로 라우팅

`swarms`에서 `TreeAgent`, `Tree`, `ForestSwarm` 임포트 후 트리 생성

Forest는 여러 트리를 가지며, 각 트리는 여러 에이전트를 가짐

Forest는 모든 에이전트를 포함하며, 트리 내부에 특화된 에이전트 존재

의료 및 금융 분야 ForestSwarm 활용 예시

금융 관련 작업을 위한 Swarm 예시 제시

의료 및 건강 관리 작업을 위한 특화된 에이전트 생성 가능성 언급

의료 분야 특화 Forest 생성 및 다양한 전문 에이전트 구성 예시

의료 관련 질문을 통해 에이전트 협업 및 작업 라우팅 과정 시연

두통, 시각 장애, 목 경직 증상에 대한 진단 및 치료 계획 요청

금융 거래 특화 Forest 생성 및 다양한 전문 에이전트 구성 예시

AI 섹터 ETF 분석 및 위험 조정 포트폴리오 배분 전략 요청

`pip install das-capital-u swarms` 명령어를 사용한 설치 방법 설명

환경 변수 설정 (workspace directory) 및 모델 이름 설정 (GPT-4, Claude 2 등)

OpenAI 또는 Anthropic API 키 설정 방법 설명

AI 섹터 ETF 분석 작업 실행 및 결과 확인

여러 Forest 결합을 통한 확장 가능성 언급

회사 전체 마케팅 활용 가능성 제시

여러 Forest를 결합하는 '메가-포레스트' 개념 언급

에이전트 간 동적 연결 및 작업 체이닝 방식 연구 계획 언급

추천 AI 섹터 ETF (BOTZ, iShares Robotics and Artificial Intelligence Multisector ETF 등) 및 구매처 소개

위험 관리 및 포트폴리오 조정 전략 제시

샘플 포트폴리오 배분 전략 제시

ForestSwarm의 동적 작업 라우팅 방식 설명

에이전트 간 공유 메모리 시스템을 통한 협업 방식 설명

Swarms GitHub, Discord 커뮤니티 참여 안내