Thanks for this article! I'm very much interested in souvereign AI and how to build an agentic AI system with own hardware, an open source framework for the agent and models hostet in the EU or even local models. I want to have the advantage of AI without being dependend on the big AI companies. Have you build something like this already?
Thank you, Bianca – I am very aligned with this instinct.
For me, sovereign AI at an individual or org level means four pieces: data under your control, compute you can trust (own hardware or a truly local cloud), open or open‑weight models that can be inspected and moved, and a governance layer that reflects EU law and your own values rather than a US platform’s defaults.
I am still in the process of assembling my own agentic stack along those lines – experimenting with open‑source agent frameworks and EU‑hosted or local models, and mapping which parts must stay on my side of the boundary and which can safely sit with a provider.
Once I have something stable enough to recommend, I will definitely write up the architecture and trade‑offs in TechLetter; it is a journey a lot of us want to make, for exactly the reasons you describe.
I'd be curious what you think about ChangeAgentAI and whether or not it meets those requirements? I am very appreciative of this post as it outlined the risks so clearly. I'm curious if the risks are really focused on using Dispatch with Claude Cowork rather than just Cowork with narrow permissions? I had thought Cowork was sandboxed and local, but Dispatch seems to open up the sandbox. Is that correct?
Thanks for this article! I'm very much interested in souvereign AI and how to build an agentic AI system with own hardware, an open source framework for the agent and models hostet in the EU or even local models. I want to have the advantage of AI without being dependend on the big AI companies. Have you build something like this already?
Thank you, Bianca – I am very aligned with this instinct.
For me, sovereign AI at an individual or org level means four pieces: data under your control, compute you can trust (own hardware or a truly local cloud), open or open‑weight models that can be inspected and moved, and a governance layer that reflects EU law and your own values rather than a US platform’s defaults.
I am still in the process of assembling my own agentic stack along those lines – experimenting with open‑source agent frameworks and EU‑hosted or local models, and mapping which parts must stay on my side of the boundary and which can safely sit with a provider.
Once I have something stable enough to recommend, I will definitely write up the architecture and trade‑offs in TechLetter; it is a journey a lot of us want to make, for exactly the reasons you describe.
I'd be curious what you think about ChangeAgentAI and whether or not it meets those requirements? I am very appreciative of this post as it outlined the risks so clearly. I'm curious if the risks are really focused on using Dispatch with Claude Cowork rather than just Cowork with narrow permissions? I had thought Cowork was sandboxed and local, but Dispatch seems to open up the sandbox. Is that correct?