Single Agent vs Multi-Agent: When to Construct a Multi-Agent System

AI Brokers

When constructing an AI agent, the design selection issues. A single agent could also be sufficient for simple duties, whereas extra advanced workflows might have a number of specialised brokers working collectively, with every one chargeable for a particular a part of the method, akin to retrieval, writing, verification, coding, testing or overview.

This publish explains the core parts of AI agent design, the ReAct method, the distinction between single-agent and multi-agent architectures, and the way to decide on the fitting design relying on the duty. It additionally features a walkthrough of how a sensible Multi-Agent RAG system works and the way it was constructed.

Inference Scaling (Check-Time Compute): Why Reasoning Fashions Elevate Your Compute Invoice

CSPNet Paper Walkthrough: Simply Higher, No Tradeoffs

widespread as a result of fashionable LLMs are actually extremely succesful at duties like coding, writing, reasoning, and fixing issues throughout totally different fields. This has lowered the necessity to practice customized fashions and shifted extra consideration towards constructing sensible functions round current LLMs. Instruments like Codex, Claude Code, Cursor and Windsurf are already serving to software program engineers work quicker, whereas companies use brokers for buyer assist, automation and different real-world duties.

An AI agent is an utility that makes use of an LLM to purpose, plan and use instruments to carry out duties, permitting the mannequin to work together with its setting in a sensible and helpful means.

Parts of an AI Agent

A number of the main parts of most AI brokers are the LLM, instruments, and reminiscence.

LLM: That is the mind of the AI agent. It’s the massive language mannequin that allows the agent to purpose, plan, and determine the best way to resolve a given process.
Instruments: These are helpers, normally within the type of code capabilities, that enable the LLM to work together with its setting. Instruments assist the agent hook up with exterior knowledge sources, search the web, retrieve info from databases, entry information, and perform particular actions. For instance, coding brokers can use instruments to jot down, debug, and save information, analysis brokers can use net search or vector databases to assemble info and buyer assist brokers can use inside firm paperwork to reply questions based mostly on trusted enterprise information.
Reminiscence: This permits the agent to retailer related info from interactions and use it later to supply higher and extra constant help. It helps the agent keep context throughout duties and enhance the general consumer expertise.Reminiscence could also be non-compulsory throughout early growth, however it turns into an necessary a part of many real-world AI agent techniques, particularly when the agent must deal with follow-up questions, multi-step workflows or personalised interactions. There are two main forms of reminiscence generally utilized in AI brokers: short-term reminiscence and long-term reminiscence. Brief-term reminiscence retains monitor of knowledge inside the present session or process, whereas long-term reminiscence shops helpful info throughout a number of periods or chats so the agent can use it later.

ReAct (Reasoning + Appearing) in Brokers

An AI agent differs from a fundamental chatbot as a result of a chatbot normally follows a extra direct workflow: consumer question → LLM → response. The LLM receives the consumer’s message and generates a reply based mostly primarily on the immediate and its current context.

An AI agent goes past this through the use of the LLM to purpose in regards to the process, determine what must be achieved, select whether or not instruments are wanted, name these instruments, observe the outcomes and proceed till it will probably produce a helpful reply.

That is the place the ReAct method is available in. ReAct means Reasoning + Appearing. It’s an agent sample the place the LLM causes a couple of process and takes actions, normally via instruments, based mostly on that reasoning. It entails designing a core logic loop round an LLM.

A fundamental ReAct workflow in an AI agent normally appears to be like like this:

Step 1: The agent receives a consumer question

The LLM causes over the duty and decides whether or not it will probably reply straight or wants to make use of instruments. It checks what instruments can be found and decides which of them are wanted to unravel the duty.

Step 2: The agent calls the required instruments

Based mostly on its reasoning, the agent takes motion by calling the mandatory instruments. These instruments might search the net, retrieve paperwork from a vector database, entry information, run code or hook up with an exterior API. The outcomes returned from these instruments are often called software outputs.

Step 3: The software outputs are despatched again to the LLM

The software outputs are handed again to the LLM as further context. This provides the agent extra related info to work with as an alternative of relying solely on the unique immediate.

Step 4: The LLM checks the proof and generates a response

The LLM critiques the software outputs and checks whether or not they’re sufficient to unravel the duty. If the proof is ample, it generates a grounded response for the consumer. If not, the agent might repeat the reasoning, tool-calling and commentary steps till it has sufficient info to supply a helpful reply.

Construction of AI Brokers

AI Brokers can both be single or multi relying on the design construction.

Single Agent vs Multi-Agent

A single agent is an agent design the place one LLM handles the entire process. It causes, plans and calls the required instruments when wanted. Most AI brokers begin as single-agent techniques as a result of they’re less complicated, simpler to take care of and normally sufficient for a lot of duties.

A multi-agent system makes use of specialised brokers to unravel totally different elements of a process. It typically has a central agent, normally known as an orchestrator, supervisor or planner, that coordinates the opposite brokers and decides when every one ought to act. Every specialised agent can have its personal position, instruments and reasoning logic, making the system extra modular and appropriate for advanced workflows.

When to Construct A Multi-Agent System

A single-agent design works nicely for easy duties that require restricted software use. For instance, a private assistant agent that may entry your calendar to guide reminders, a calculator agent that solely makes use of a calculator software, or an internet search agent that makes use of an internet search API to retrieve up-to-date info.

Nonetheless, a single agent can change into overloaded when the duty requires many instruments, multi-step reasoning, totally different duties or verification earlier than the ultimate response is returned to the consumer. Widespread points embody overloaded prompting, poor software routing, unclear agent duties and lowered reliability as a consequence of an excessive amount of complexity in a single agent.

A multi-agent system is a more sensible choice when the duty might overwhelm a single-agent design and if you want specialised brokers with clear roles, their very own instruments and separate duties.

For instance, a software program engineering agent may go higher as a multi-agent system:

Orchestrator → Coder → Tester → Reviewer

The Orchestrator coordinates the workflow, the Coder agent generates the code, the Tester agent checks whether or not the code works, and the Reviewer agent critiques the answer to verify for lacking elements or potential enhancements.

One other instance is a analysis agent that researches a subject, retrieves info from totally different knowledge sources and generates grounded content material:

Orchestrator → Retriever → Author → Verifier

The Retriever agent gathers info from the net and native paperwork saved in a vector database. The Author agent writes based mostly on the retrieved content material. The Verifier agent checks the written content material for errors, citations and factual accuracy earlier than the ultimate response is returned.

Multi-agent techniques make the workflow extra modular and provides every stage a transparent position. Nonetheless, they need to be used solely when the duty genuinely wants that design, as a result of they normally enhance latency, value and upkeep complexity as a consequence of extra LLM calls and extra transferring elements.

A easy rule is:

Use a single agent when the duty is easy, has fewer steps and desires only some instruments. Use a multi-agent system when the duty requires specialised roles, multi-step reasoning, stronger verification or coordination throughout totally different instruments and workflows.

Walkthrough of A Multi-Agent Challenge

I constructed a mission known as Multi-Agent RAG Researcher to make the thought of multi-agent techniques extra sensible.

The purpose of the mission is to indicate how a central agent can coordinate a number of specialised brokers to analysis a subject, retrieve proof from paperwork and the net, write a grounded content material and confirm the content material earlier than returning it to the consumer. As a substitute of utilizing one agent to deal with every part, the system splits the workflow into totally different duties.

Examine the mission on github: https://github.com/ayoolaolafenwa/multi-agent-rag-researcher

Clone Challenge repo

git clone https://github.com/ayoolaolafenwa/multi-agent-rag-researcher.git

Clone the repo to followup with the code alongside the publish. When the repo is cloned, the mission construction will appear like this:

.
├── docs/                         # Default PDF information
├── reminiscence/                       # SQLite-backed session reminiscence helpers
├── qdrant_vector_database/       # PDF ingestion and similarity search
├── ui/                           # Gradio app and UI handlers
├── utils/
│   ├── necessities.txt          # Python dependencies
├── worker_agents/                # Retriever, author, and verifier
├── orchestrator_agent.py         # Major coordinator
└── run_orchestrator.py           # CLI entry level

Multi-Agent Structure

Information Sources

There are two main knowledge sources:

Qdrant Vector Database

Data retrieval from PDFs is dealt with within the following levels:

A number of PDFs will be loaded from the docs/ folder or uploaded via the UI.
Paperwork are break up into chunks, transformed into embeddings, and saved in a neighborhood Qdrant assortment.
Similarity search is then used to retrieve essentially the most related chunks throughout the listed paperwork.
The retrieved chunks embody quotation metadata akin to doc identify and web page quantity.

The doc retrieval a part of the mission the place Qdrant vector database is setup, PDF ingestion, chunking, embedding, and similarity search are managed is dealt with in qdrant_vector_database/vector_store.py .

Tavily Internet Search

Tavily is used to retrieve up-to-date or exterior info from the net. The retriever agent can use it when:

the listed PDFs don’t cowl the question
doc proof is weak or incomplete
newer info is required

Employee Brokers

Retriever Agent

The position is:

It makes use of two instruments: PDF doc retrieval and net search.
Given a question, it decides whether or not to make use of native paperwork, net search or each.
If native doc proof is lacking or weak, it will probably fall again to net search to assemble broader or extra up-to-date context.

The code for the retriever agent with tavily net search out there in worker_agents/retriever.py . It makes use of gpt-5.4-mini with low reasoning effort.

Author Agent

The position is:

It receives the retrieved info from the Retriever Agent.
It writes a grounded draft based mostly on the out there proof.
It contains supporting citations from PDFs or net sources when they’re out there.

The code for the author agent out there in worker_agents/author.py . It makes use of gpt-5.4 with low reasoning effort.

Verifier Agent

The position is:

It receives the draft from the Author Agent along with the proof.
It checks whether or not the claims within the draft are supported by the retrieved proof.
It returns the ultimate verified response.

The code for the employee agent is on the market in worker_agents/verifier.py . It makes use of gpt-5.4 with low reasoning effort.

Reminiscence

SQLite is used to supply short-term reminiscence for the multi-agent workflow. For a given session ID, the system shops:

the most recent consumer question
the most recent retrieved proof for that session

This permits the orchestrator to reuse related proof for follow-up questions as an alternative of retrieving the identical info once more each time.

The code for the reminiscence is on the market in reminiscence/reminiscence.py .

Orchestrator

The orchestrator coordinates the three employee brokers: Retriever, Author and Verifier.

How the Orchestrator coordinates the Multi-Agent Workflow

It receives the consumer question and, relying on the question, might reply straight or start the evidence-based workflow.
For a analysis question, it first checks whether or not related cached proof from the reminiscence for the present session will be reused.
If cached proof shouldn’t be sufficient, it calls the Retriever Agent to assemble proof from PDFs, the net or each.
If there’s doc proof however the proof is weak, the Retriever Agent also can fetch up-to-date info from the net to complement the native doc info.
The orchestrator then passes the lively proof and the consumer question to the Author Agent so it will probably generate a grounded draft.
Subsequent, it sends the draft and proof to the Verifier Agent, which checks the claims and returns the ultimate verified report.
Through the session, the most recent question and retrieved proof are saved in reminiscence for follow-up questions.
In follow-up questions, the orchestrator might reuse cached proof as an alternative of calling the Retriever Agent once more, then proceed with the Author Agent and Verifier Agent to generate the ultimate response.

The code for the orchestrator is in orchestrator_agent.py . It makes use of gpt-5.4-mini with low reasoning effort.

The orchestrator has a guardrail that retains the system centered on analysis and factual questions. It refuses unrelated common duties akin to coding assist or simple arithmetic as a result of the purpose of the system is to operate as a analysis assistant.

Observe: For the fashions used within the orchestrator and employee brokers, you’ll be able to change them from gpt-5.4 to any openai supplied mannequin of your selection.

Challenge Setup

Conditions

Python 3.10 or newer
OpenAI API key: Create an OpenAI Account when you don’t have one and Generate an API Key.
Tavily API key: Tavily is a specialised web-search software for AI brokers. Create an account on Tavily.com, as soon as your profile is ready up, an API key can be generated that you could copy into your setting. New account receives 1000 free credit that can be utilized for as much as 1000 net searches.

Set up

Create and activate a digital setting:

python3 -m venv env
supply env/bin/activate

2. Set up the dependencies:

cd multi-agent-rag-researcher
pip3 set up -r utils/necessities.txt

3. Create a utils/var.env file and retailer your API keys:

OPENAI_API_KEY=your_openai_api_key
TAVILY_API_KEY=your_tavily_api_key

4. Place the PDFs you wish to index within the docs/ folder, or add PDFs later via the UI. The mission already contains current PDFs in docs/, at present Gemma 3 Technical Report.pdf and DeepSeek-V3.2.pdf, so you should use these straight or substitute them with your individual paperwork.

Run Challenge

Begin the command-line app:

python3 run_orchestrator.py

When the CLI begins, it ingests the PDFs in docs/ into the native Qdrant retailer. Sort q or exit to finish the session.

Run UI for Multi-Agent Chat

Begin the Gradio UI:

python3 ui/gradio_app.py

The UI mechanically masses the default PDFs from docs/ on startup. When you add new PDFs, they substitute the lively listed doc set for that UI session.

Demo Video of the Multi Agent Agent RAG Researcher

Notes

Session reminiscence is saved in utils/reminiscence.db.
Native Qdrant knowledge is saved in utils/qdrant_storage/.
The system is designed for analysis and factual query answering, not for unrelated general-purpose duties.

Conclusion

On this publish, I defined how an AI agent works, the way it makes use of instruments to work together with its setting, and the way the ReAct method helps it purpose, plan, choose instruments and execute particular duties.

I additionally lined the structural design of AI brokers, which will be single-agent or multi-agent techniques. I defined how each designs work, when to decide on every one based mostly on the workflow, and in contrast single-agent implementation with multi-agent structure.

Lastly, I did a walkthrough of the multi-agent design behind my Multi-Agent RAG Researcher mission, exhibiting the way it makes use of an orchestrator to coordinate three employee brokers, retrieve info from the net and native paperwork, use reminiscence for consistency and write and confirm grounded content material earlier than returning the ultimate output.