Build AI agent without code: wetin you go use

How to build AI agent without code: the short answer

If you wan build AI agent without code, you no need learn Python. You go install one visual builder for your own VPS, join boxes for screen, connect one AI model, and the agent go start to answer people. For beginner, we recommend Langflow. E dey run as one Docker container, and e carry MIT licence, so you fit use am for paying client work.

This guide compare the four builders wey people dey host for their own server: n8n, Dify, Langflow and Flowise. We look the things wey you go really decide on: how you go assemble agent for each one, how big server e need, the licence if you dey build for clients, and who still dey maintain am. After that, we build one first agent for Langflow 1.12.5, step by step.

Who this guide dey for

You fit be founder for Lagos wey wan make customers get answer for night. You fit be student wey wan show working project. You fit run small school wey wan sort student documents. You fit get shop for Aba wey tire to answer "how much be this one?" fifty times every day. You no wan write code. You want one agent wey sabi your business and wey no go cost you plenty dollar every month.

AI agent na program wey use LLM (large language model) to decide wetin e go do next. E fit call tool before e reply, like to check your price list or read one web page. Plain chatbot just reply from wetin the model already sabi. If the difference never clear, read the difference between AI agent, LLM and assistant first. Then the different types of AI agents go show you which kind you really need.

The four no-code AI agent builders wey you fit host yourself

• n8n: workflow automation tool. You join nodes for canvas, and AI na one part of the workflow. Its GitHub page call am "Fair-code workflow automation platform with native AI capabilities."
• Dify: platform for building AI apps. Agent na one type of app inside am.
• Langflow: visual builder for agents and flows. You drag components put for canvas and connect their ports.
• Flowise: visual builder wey many people don use before. The project don shut down. Read the maintenance section before you touch am.

How each tool dey assemble agent

n8n: the agent na one step inside workflow

For n8n, the agent na one step inside bigger workflow. Something start the workflow, like new email. The agent step decide wetin to do. Then other nodes carry the result go where e suppose go. This one strong when your agent need touch plenty other apps. We no go teach n8n again here, because our guide to build n8n AI agent on your own VPS don cover am from start to finish.

Dify: the agent na one app

For Dify, you create Agent app. You write instructions wey tell the agent how e go behave. You add tools, so e fit reach other services and APIs. If you want make e answer from your own documents and no be from general knowledge, you attach knowledge base. Then you pick model wey support tool calling, and you click Publish. If the model no support tool calling by itself, Dify go run am for ReAct mode, where the model write out im reasoning before e call tool.

That knowledge base na strong reason to pick Dify if your main job na to answer from documents. Our guide to self-host Dify on a VPS show the full install.

Langflow: the agent na one component for canvas

For Langflow, you put Agent component for canvas, add Chat Input and Chat Output, and connect dem. You choose model provider under Settings > Model Providers. To give the agent tool, you turn on Tool Mode for one tool component, then you connect its Toolset port to the Tools port for the Agent. You test everything inside Playground. Na this one we go build below.

Flowise: no new build

We no go show Flowise steps. The project don reach end of life, as the maintenance section explain. Beginner no suppose start new work on top project wey nobody dey fix again.

How much server each no-code builder need

These numbers come from each project own docs. Dem be minimum or example figures. We no measure dem, and your own usage fit pass dem.

• Dify: at least 2 CPU cores and 4 GiB RAM, plus Docker Compose 2.24.0 or later. Dify no be one container. Im compose file start API, worker, web, PostgreSQL, Redis, sandbox, plugin daemon, Nginx and vector store together.
• Langflow: minimum na dual-core CPU and 2 GB RAM. Dem recommend multi-core CPU with at least 4 GB RAM.
• n8n: the n8n sizing guide for OEM deployments, wey base on n8n Cloud, give memory from 320 MB to 2 GB. The same page say n8n Cloud instance wey dey idle need about 100 MB, and say n8n no dey use plenty CPU.
• Flowise: docs say Node 18.15.0 or 20 and above, and e dey run for port 3000. We no see RAM figure for Flowise docs.

Docs figure no be your figure. After you don use your builder small, run these two commands for the server:

free -h
sudo docker stats --no-stream

free -h show how much RAM the whole server still get. docker stats show how much each container dey use. If you plan to run the AI model for the same server with Ollama, the model need RAM on top of all this. We no find RAM figure per model for Ollama FAQ or README, so check the model page for the Ollama library before you choose VPS size.

Licence: wetin matter if you dey build AI agents for clients

This part na summary of the licence text. E no be legal advice. If money dey involved, read the licence yourself.

n8n: Sustainable Use License

n8n use Sustainable Use License, version 1.0. E no be OSI open source licence. n8n call am "fair-code". The licence say you fit use or modify n8n only for your own internal business, or for non-commercial or personal use. You fit give am to other people only if you do am free and no be for commercial purpose.

The n8n licence FAQ clear some client cases. You fit build automations for your clients on top your own instance, as long as the clients no fit create or edit dem. You fit charge clients for workflow creation, setup and maintenance. But you no fit host n8n as service where clients go build their own workflows. Files wey get .ee for their name or folder no dey under this licence. Dem need separate n8n Enterprise License. If your case no clear, email license@n8n.io. For the full gist, read our breakdown of the n8n Sustainable Use License and wetin free for n8n Community and wetin need Enterprise.

Dify: modified Apache 2.0

Dify use modified Apache 2.0 licence wey dem call "Dify Open Source License". E add two extra conditions. First one: you no fit run multi-tenant service unless Dify give you permission for writing. For Dify, one tenant na one workspace. Second one: you no fit remove or change the logo or copyright information for the frontend, wey be the web folder or the web Docker image.

So the case wey you must check na agency wey give each client im own workspace for one shared Dify install. By the licence text, that one na multi-tenant. If you dey plan separate Dify install for each client, the licence text no talk about that case directly. Read the Dify LICENSE yourself, or ask Dify. The licence say commercial licence must come from the producer.

The Dify pricing page describe the free Community self-hosted plan as "For open-source enthusiasts, individual developers, and non-commercial projects." E list "Commercial License Authorization" under Enterprise. The licence text itself no ban commercial use outright, but the two pages no read the same way. If you dey build for paying client, read both before you sign contract.

Langflow and Flowise: MIT and Apache 2.0

Langflow use MIT licence. You fit use am for client work. If you share the code, keep the copyright notice. Flowise use Apache 2.0, except one enterprise folder, packages/server/src/enterprise, wey dey under commercial licence.

Who dey maintain each project today

• n8n: the n8n-io organisation for GitHub. When we check on 9 October 2026, latest release na n8n@2.42.6, wey come out that same day.
• Dify: LangGenius, under the langgenius organisation for GitHub. Latest release na 1.17.1, from 10 September 2026. The repo still dey get new commits.
• Langflow: the langflow-ai organisation for GitHub. Latest release na v1.12.5, from 6 October 2026. For February 2025, IBM announce say e go buy DataStax, the company wey own Langflow.
• Flowise: nobody, officially. Workday announce for August 2025 say e don buy Flowise.

For Flowise, the team announce code freeze on 29 July 2026 and end of life on 31 August 2026. The GitHub repo don turn public archive since August 2026, so e be read-only now. The last release na flowise@3.1.4, from 29 July 2026. The npm packages and Docker images dey marked deprecated. The Flowise sunset page say the Apache 2.0 code still dey GitHub, so person fit fork am. But nobody go send official security fix again, so no put new customer agent on top am.

Hosted builder or your own VPS: how the money dey move

Hosted builders dey, and dem dey work. You no need server, you just pay subscription. But most of dem bill for dollar or euro, so exchange rate go touch your bill every month. These na the prices wey the pricing pages show for October 2026. Check am again before you pay.

• Zapier bill for US dollars, by tasks per month. Free plan get 100 tasks per month. Pro na $19.99 per month for 750 tasks if you pay for the year, or $29.99 if you pay month by month.
• n8n Cloud bill for euro, by workflow executions. Starter na 20€ per month and Pro na 50€ per month, billed yearly. Starter get 2.5K workflow executions. The trial no need credit card.
• Dify Cloud bill per workspace per year, plus message credits for model use. Sandbox free with 200 credits. Professional na $590 per year with 5,000 credits per month. When credits finish, you fit switch to your own API key.

When you host am yourself, the cost get two parts. The first na flat VPS price every month, no matter how many messages pass. The second na model tokens. Token na small piece of text, and hosted model providers dey charge you per token wey your agent send and receive. If you run local model with Ollama, the token cost comot. You pay with RAM instead, and small model wey fit inside VPS RAM fit give weaker answers. Test am with your real customer questions before you trust am.

Walkthrough: build your first AI agent for Langflow 1.12.5

We go build customer assistant for small fabric shop. E go answer questions about price and delivery, and tell customers when the shop dey open. We assume Ubuntu 24.04 VPS with at least 4 GB RAM (the Langflow recommendation). You need user wey get sudo and SSH access from your laptop. You go also need API key from one model provider, because Langflow no come with model inside.

Step 1: Install Docker

sudo apt update
sudo apt install -y docker.io openssl curl
sudo systemctl enable --now docker
sudo docker run --rm hello-world

The last command must print Hello from Docker!. If e show permission denied near docker.sock, you forget sudo.

Step 2: Put the admin password inside one safe file

sudo install -d -m 700 /etc/langflow
sudo sh -c 'umask 077; printf "LANGFLOW_AUTO_LOGIN=false\nLANGFLOW_SUPERUSER_PASSWORD=%s\nLANGFLOW_CONFIG_DIR=/app/langflow\n" "$(openssl rand -base64 24)" > /etc/langflow/langflow.env'
sudo cat /etc/langflow/langflow.env

Copy the password from the output and keep am for your password manager. Only root fit read this file.

LANGFLOW_AUTO_LOGIN=false mean say everybody must log in. The Langflow application default na True, wey mean say anybody wey open the page enter as superuser. Official Docker images already set am to false, but we write am inside the file so nobody go guess. When auto login dey off, LANGFLOW_SUPERUSER_PASSWORD must dey. The username default na langflow. The password no fit be langflow, because Langflow go refuse to start.

We keep the password for file and no be for shell variable. Shell variable go disappear when you log out, and you go need the same settings again when you upgrade.

Step 3: Start Langflow with pinned image tag

sudo docker run -d --name langflow --restart unless-stopped \
  -p 127.0.0.1:7860:7860 \
  --env-file /etc/langflow/langflow.env \
  -v langflow-data:/app/langflow \
  langflowai/langflow:1.12.5

Each line get im reason. -p 127.0.0.1:7860:7860 publish the port only for the server itself, so nobody for internet fit reach the login page. -v langflow-data:/app/langflow keep Langflow config and data inside Docker volume, outside the container. Langflow docs advise am, because when you upgrade, you go replace only the container image. The tag 1.12.5 dey pinned, so the version no go change until you change am yourself. Langflow use SQLite database by default. If you later want PostgreSQL, the setting na LANGFLOW_DATABASE_URL.

Step 4: Check say Langflow dey run

sudo docker ps --filter name=langflow
curl -s -o /dev/null -w '%{http_code}\n' http://127.0.0.1:7860/

docker ps must show the langflow container with status Up. The curl line must print 200. First start fit take some time, so if e print 000, wait small and run am again. 000 mean say nothing answer for port 7860 yet.

Step 5: Open Langflow from your laptop with SSH tunnel

The port no dey open to internet, so you go reach am through SSH. Run this one for your laptop, no be for the server:

ssh -L 7860:127.0.0.1:7860 your_user@your_server_ip

Leave that SSH window open. Open http://localhost:7860 for your browser. Log in with username langflow and the password from Step 2. The tunnel live only while that SSH session dey open. When you close am, the page stop to load.

Step 6: Build the agent for canvas

1. Create new blank flow.
2. Add Agent component to the canvas.
3. Add Chat Input and Chat Output components.
4. Connect Chat Input to the Agent input, and connect the Agent output to Chat Output.
5. Open Settings > Model Providers, choose your provider, and paste your API key.
6. Inside the Agent component, choose the model wey you wan use. Langflow no pick model for you.
7. Write the instructions for the Agent.

For instructions, write the rules of your business plainly. Example:

You are the customer assistant for Ada Fabrics, a small shop in Yaba, Lagos.
Answer only questions about our fabrics, prices, delivery and opening hours.
Ankara is sold by the yard. Delivery inside Lagos takes 1 to 2 days.
We open Monday to Saturday, 9am to 6pm.
If you do not know an answer, say so and give the shop phone number: 0800 000 0000.
Never promise a discount.

If your customers dey write Pidgin, tell the agent make e reply for Pidgin. Open Playground and ask am "Una dey open for Sunday?" The agent must answer from your rules. If e invent answer wey no dey your rules, make the instructions more strict.

Step 7: Give the agent one tool

Agent without tool na just chatbot with instructions. Pick one tool component wey your agent need, like component wey fit read web page where your price list dey. Turn on Tool Mode for that component. Connect im Toolset port to the Tools port for the Agent. Ask question for Playground wey need that tool, and check say the agent use am.

Tools wey only read information dey safe to start with. Tools wey fit send message, change order or spend money need human to approve first. How to gate AI agent actions with human approval explain why and how.

Use Ollama to remove the token cost

Ollama run open models for your own server. Install command for Linux, from the Ollama README:

curl -fsSL https://ollama.com/install.sh | sh

By default, Ollama listen on 127.0.0.1 port 11434. Langflow get Ollama model component, and im default base URL na http://127.0.0.1:11434. Here na the trap. Langflow dey inside container, so 127.0.0.1 inside the container na the container itself, no be your server. The default URL go point to place wey Ollama no dey. You go need make Ollama listen on address wey the container fit reach (the OLLAMA_HOST setting change the bind address), then put that address for the Base URL. We never test one exact setup for this guide, so follow Ollama and Docker docs for your case. No set OLLAMA_HOST to 0.0.0.0 on server wey firewall no protect, because that one open your model to the whole internet.

Also, we no confirm say every Ollama model fit call tools inside Langflow Agent. Test your model for Playground with one tool before you build on top am.

WhatsApp or Telegram: how customers go reach your agent

Most Nigerian customers dey WhatsApp, so e make sense to want am there. But WhatsApp get process and cost. You need business portfolio to use the WhatsApp Business Platform, plus Facebook account and developer registration to start. Template messages generally need approval before you fit send dem, and you must get user opt-in first.

Meta charge per message wey e deliver. Messages dey four categories: marketing, utility, authentication and service. When customer send you message, e open 24-hour customer service window, and service messages inside that window free. Messages wey come from click-to-WhatsApp ads or Facebook Page buttons get 72-hour free window. Nigeria get im own line for the rate card. Rates dey change, so check Meta WhatsApp pricing on the day you onboard.

For demo, Telegram easier. Message @BotFather for Telegram to register your bot, and e go give you bot token. We no check whether Langflow get Telegram component. Search Langflow docs for "Telegram" by name before you plan. If no component dey, you go need small bridge wey receive Telegram messages and pass dem to your flow, and that bridge na code.

When e break: the errors wey you go see

docker ps no show langflow. The container start and die. Run sudo docker ps -a --filter name=langflow, and you go see Exited. Then run sudo docker logs langflow to read why. Common cause na password wey empty or wey be langflow, because Langflow refuse that legacy password.

curl print 000. Nothing dey answer for port 7860. Either Langflow still dey start, or the container don stop. Check docker ps first.

SSH print Could not request local forwarding. Something for your laptop already use port 7860. Use another local port, like ssh -L 8860:127.0.0.1:7860 your_user@your_server_ip, then open http://localhost:8860.

Browser no fit load localhost:7860. Your SSH tunnel don close. Open the ssh -L command again.

Server slow, or container restart by itself. RAM don finish. Run free -h. Run sudo dmesg | grep -i "out of memory". If you see Out of memory: Killed process, the kernel kill one process to free RAM. Move to bigger VPS, or comot Ollama from the same box.

Your flows disappear after you remove the container. You run the container without -v langflow-data:/app/langflow. Without volume, the data live inside the container, and docker rm delete am together.

How to upgrade Langflow and keep your flows

First, stop Langflow and back up the volume:

sudo docker stop langflow
sudo docker run --rm -v langflow-data:/data -v "$PWD":/backup alpine tar czf /backup/langflow-data.tgz -C /data .

Check the new release for the Langflow GitHub releases page and the tag for Docker Hub. The Docker tag no get v in front. Then replace only the container, with the same volume and the same env file:

sudo docker pull langflowai/langflow:NEW_TAG
sudo docker rm langflow
sudo docker run -d --name langflow --restart unless-stopped \
  -p 127.0.0.1:7860:7860 \
  --env-file /etc/langflow/langflow.env \
  -v langflow-data:/app/langflow \
  langflowai/langflow:NEW_TAG

Change NEW_TAG to the real tag. Docker go attach the same volume, so your flows and database go still dey. Run the Step 4 checks again.

If one day you decide to write code, the same server fit carry am. Build your own AI agent on a VPS with code show that road.

FAQ

Which no-code AI agent builder good pass for beginner?

For beginner wey no wan write code, Langflow na easy start. E run as one Docker container, im docs minimum na 2 GB RAM with 4 GB recommended, and e use MIT licence. Dify better when your agent must answer from your documents, because im Agent app fit attach knowledge base, but e need at least 4 GiB RAM and plenty containers. n8n better when the agent must connect to many other apps.

Can I build AI agents for my clients with n8n or Dify and collect money?

With n8n, the licence FAQ allow you build automations for clients on your own instance, if the clients no fit create or edit dem, and e allow you charge for setup and maintenance. E no allow you host n8n as service where clients build their own workflows. With Dify, one shared install where each client get im own workspace na multi-tenant, and that one need written permission from Dify. Read each licence, and ask the vendor if your case no clear.

Flowise still dey safe to use?

The Flowise team announce end of life on 31 August 2026, and the GitHub repo don turn read-only archive since August 2026. The last release na flowise@3.1.4, and the npm packages and Docker images dey marked deprecated. The Apache 2.0 code still dey GitHub for anybody wey wan fork am, but no official fixes dey come again. For new agent, use Langflow or Dify.

How much RAM my VPS need to run Langflow?

Langflow docs give minimum of dual-core CPU and 2 GB RAM, and recommend multi-core CPU with at least 4 GB RAM. That one na for Langflow alone. If you run local model with Ollama for the same server, the model need extra RAM, so check the model page for the Ollama library. After you start, run sudo docker stats --no-stream to see wetin Langflow really dey use.

I fit connect my AI agent to WhatsApp?

Yes, through Meta WhatsApp Business Platform, but you need business portfolio, and Meta charge per message wey e deliver. Service messages inside the 24-hour window wey customer message open dey free, and template messages need approval and user opt-in. Check the rate card for Nigeria on Meta pricing page before you start. For quick demo, Telegram bot from @BotFather easier.