What “free” means for a business
Free here means one thing: you pay nothing for the program’s license. The agent’s work still costs money: a computer or server, power, and people’s time for setup and checks.
An open source license allows you to install a program, change it and use it for commercial purposes. A fair-code license (the code is open, commercial use is limited) allows part of this: n8n, for example, can be used for a company’s internal tasks, but cannot be passed to clients for commercial purposes free of charge.
So read the license first and look at the features second. For an agent inside your own company, every tool in the list below fits. If the agent becomes part of a product for your clients, check the n8n and Dify terms separately.
A typical agent is built from these open blocks. If someone charges for it as a unique development, ask what the price is made of.
Tools and licenses
n8n: an automation platform with AI agents from Berlin, around since 2019. The Sustainable Use License allows you to use and change n8n for a company’s internal tasks and for non-commercial and personal purposes. You may pass it on to others only free of charge and not for commerce.
This is not an OSI open source license. Until 17 March 2022, n8n came under Apache 2.0 with the Commons Clause added, and for use the current license does not allow, n8n offers a separate commercial agreement.
Flowise: a visual builder for AI agents, under the Apache 2.0 license, with 55 thousand stars on GitHub. You build an agent from blocks with the mouse, without code.
Dify: a platform for AI applications and agents under Apache 2.0 with two conditions. Without permission, you may not run a multi-tenant cloud service like Dify or remove the Dify logo from the console.
CrewAI: a framework (ready code a programmer uses to build their own solution) for teams of agents with roles and tasks, MIT license, first release in December 2023. LangChain and LangGraph: libraries for building agents in Python and JavaScript, also MIT. A programmer works with them, and they do not replace a no-code builder.
Ollama: a program for running open models on your computer or server, MIT license. The review of open tools compares them by task.
Models and licenses
The model writes the answers, and the tool manages the agent’s steps. Both parts can be open, and you check the license of each.
Open weights mean you can download the model file and run it on your own machine, with no subscription to someone else’s service. The model works on your hardware, and staff queries go nowhere.
DeepSeek R1 came out on 20 January 2025 with open weights under the MIT license. The Qwen 3 family came out on 29 April 2025 with open weights under the Apache 2.0 license.
The difference between the licenses is small. MIT allows almost any use if you keep the license text and the author’s name. Apache 2.0 adds an explicit grant of the author’s patents and asks you to mark which files you changed.
Both licenses allow a company to use the models in its work without paying the model developer. You pay for the hardware: the larger the model, the more memory it needs.
The page on open models covers which other open models exist, with their sizes and licenses.
Where costs will appear
The first item: hardware. Ollama runs the model on your computer or server, and its memory decides which model fits. Calculate the price of your server in the local AI cost calculator.
The second item: a cloud model, if an open one cannot handle the task. Then you pay for each request, and the bill grows with the number of requests.
The third item: a license, if the agent goes beyond internal tasks. For that case n8n offers a separate commercial agreement, and Dify requires permission for a cloud service for many clients.
The fourth item: people. Someone sets up the agent, checks answers, updates documents and looks into failures. The page on the cost of an AI agent covers what the full price of an agent is made of.
How to start
Start with one task and one tool. The order is:
1. Choose a task with a clear right answer, for example answering staff questions from company policies.
2. Install Ollama and download an open model: Qwen 3 or DeepSeek R1.
3. Build the agent in a no-code builder, n8n or Flowise. A step-by-step guide on n8n is on the page on a first agent in one evening.
4. Check the answers on your own questions, and only then give the agent to colleagues.
5. Write down how many hours a week the agent saves. Without this number, the decision on the next agent is a guess.
The review of do-it-yourself agents covers what you can really build without a programmer and where you need one.