You Should Keep Sensitive AI Workflows Open Source

16 Aug 2026 · 1 min

Watch on YouTube

0:34

In this brief commentary, the speaker expresses a lack of trust in Anthropic and OpenAI for handling highly sensitive tasks, citing concerns about corporate incentives and executive behavior. He promotes the newly released open‑source Hermes Voice model, highlighting that it can be run locally on personal hardware and activated like a voice assistant (e.g., “Hey Hermes”). He explains that his Hermes harness, combined with OpenWeights, enables powerful, self‑hosted inference. While warning about the motivations of large AI companies, he admits he still leverages their services when needed, underscoring a pragmatic but cautious stance toward proprietary AI.

Chapters

I don't trust Anthropic or OpenAI. I wouldn't go like with super sensitive stuff. Ultimately, here's what I would say if you want to run this, because Hermes Voice just came out yesterday, which is open source, by the way. So Hermes, you can run on your own box, you can run on your own systems, you can talk to it, you can activate it just saying, hey Hermes, kind of like, hey Siri. And because my Hermes harness is very powerful, I can just run it through that and you can run it through OpenWeights, like open source model. You can definitely do that. Just keep in mind that they have their own incentives. Ultimately, I think there's also good reason to not trust both companies because of how both of the— let's just say the executives, how they operate. And I believe that how you do anything is how you do everything. That means That being said, I still take advantage of it.