AI tools are often described as “open source”. That can refer to the application code, the model generating answers or simply the option to install something yourself. Before using it, you need to know what you actually get.
Start with the terms
Open-source software allows you to inspect, use, modify and distribute its source code under the licence terms. The Open Source Initiative’s definition goes beyond code being publicly available. A public project can still impose restrictions that matter for your intended use.
With AI, a model may be downloadable without all the information used to build it. “Open weights” refers to the available learned parameters. That alone does not make an entire AI system open source. The OSI has a separate definition for AI.
What does this mean in practice?
I look at the choices you retain. Can you export your data? Can someone else take over maintenance? Are you allowed to adapt the software for your business? Those are more useful questions than whether something calls itself open.
Self-hosting can provide more control, but someone still has to maintain the system. Even without licence fees, you need time, computing resources, backups and updates. An application may also call an external AI service, bringing additional costs and data transfers.
What I would check first
- Which licences apply to the software and the chosen model?
- Where are questions, files and answers processed?
- Who handles updates and problems?
- Can you export your information if you want to switch?
Keeping an eye on AI means looking beyond the label. Open source can be a good choice; configuration and maintenance determine whether it works for you.
Sources and further reading
Explanation and commentary based on the sources below. Examples are illustrative.