AI Swiss’s Open BASE Framework Aims to Let Firms Switch AI Providers Without Losing Their Methods
Can a company move from one AI provider to another without starting from scratch? Le Temps on October 26 profiled BASE, an open project from the Swiss association AI Swiss that tries to make the answer yes. Created by Charles-Edouard Bardyn, a researcher and co-founder of AI Swiss who serves as its scientific director, BASE is not a new Swiss model. It is a framework for organising the information and procedures a team relies on so that different models can use them, without locking that know-how inside one vendor’s interface.
The risk it targets is familiar. Le Temps notes that organisations hand documents and procedures to ChatGPT, Claude or Gemini and gradually reorganise their work around them, so a price rise or a change of terms can make migration painful. BASE keeps the rules, roles, sources, decision criteria and verification steps for generative AI work in editable text files that stay independent of models and platforms, ICTjournal reported when the framework was released in June. The files can be read, shared and adapted without programming skills.
Version 1.0.0 is on GitHub under the Apache 2.0 licence for code and CC BY 4.0 for documentation, with resources in French and English and about a dozen business examples. It includes a file structure that separates instructions from sources, a routing mechanism for requests and BASE Studio, a local application for viewing and maintaining the content, with evaluations planned to track process quality. AI Swiss’s example is someone who regularly drafts the same kind of document, such as a quote, an offer or a report, and wants its requirements, criteria and checks written down once.
What it means in practice
For Swiss SMEs and public offices, BASE is a low-cost hedge against vendor lock-in that fits a wider sovereignty debate, from warnings that Switzerland missed a chance to set its own AI vision to the Federal Council’s priority areas for AI in the federal administration. AI Swiss is explicit about limits: BASE does not replace internal governance, cybersecurity or data protection, and it does not guarantee that model answers are accurate.
Whether it becomes a real standard will depend on adoption beyond early users and on whether makers of AI tools learn to read the same files natively. For now, it gives teams a portable record of how they work with AI that they own outright.