Today, we're launching Typeless for Linux. Typeless is now available on Mac, Windows, iOS, Android — and Linux. The Linux app brings the full Typeless desktop experience, with all the core features available on Mac. But this launch is about more than adding another platform. For us, every new platform is another step toward a bigger goal: making voice the default way for people to interact with computers — wherever they think, create, and work. For decades, the keyboard has been the default interface between our thoughts and computers. And we've gotten so used to it that we rarely question the friction. Thoughts come fast, messy, and nonlinear. Before you communicate them, you normally have to turn them into sentences in your head first. Then you slow everything down again to type. We built Typeless to remove both layers of friction. With Typeless, you can just think out loud. Follow your thoughts wherever they go. Change your mind halfway through. Add context later. Repeat yourself. Speak imperfectly. Typeless turns that messy stream of thought into clear, polished writing in real time. It removes filler words and repetition, structures your ideas, adapts to your writing style, understands your personal vocabulary, and gives you writing that fits the app you're working in. You can: • Dictate — turn messy thoughts into clear writing, 4x faster than typing • Translate — speak in your language and write naturally in another • Ask anything — use your voice to write from scratch, or select text to edit it, ask questions, and take actions • Build your dictionary — names, acronyms, and specialized terms get recognized correctly • Personalize your style — Typeless learns how you write and adapts to where you're writing • Speak 100+ languages — including naturally mixing languages while you talk And it's private by design. Linux users tend to be some of the people who spend the most time interacting deeply with computers. Developers. Engineers. Researchers. Builders. They should be able to use voice as naturally on Linux as anywhere else. Our long-term belief is simple: Voice will become the default way people interact with computers. Not because speaking is merely faster than typing. But because voice gives us a way to communicate with computers much closer to the speed and form in which thoughts actually happen. To us, that's what real freedom in computing looks like: expressing what you mean without adapting how you think. And for that future to happen, it can't belong to one device, one operating system, or one workflow. It has to be available wherever you work. Linux. Mac. Windows. iOS. Android. One interface, wherever you work. Typeless for Linux is here.

We’ve just released the preview version of our latest paper: Infinite-Parameter LLMs — Generating and Adapting Weights from Live Data. In it, we demonstrate that large language models can learn up to 1,000x faster through Bayesian Self-learning transformers (BAST) than SOTA training algorithms engineered by human researchers. This breakthrough marks a critical shift in AI research, from cost-intensive, static-weight AI to energy-efficient, dynamic-weight AI. For a decade, building more advanced models has been achieved by training bigger ones on more data. This approach is now hitting a ceiling with the supply of pretraining text projected to run out in the next two to five years. Meanwhile, the fast adoption of AI agents is producing an unprecedented amount of continuously-growing data that models can learn from. None of it is captured, locked in individual sessions and lost as soon as the agent completes the task. Bringing self-learning AI agents to users captures the value of that data. We validated that BAST LLMs overcome the fundamental limitations of memory-based learning, such as Retrieval-Augmented Generation (RAG), in both cost and performance, across long-context conversations. Without underlying architectural innovation like BAST, RAG, larger context windows, and agent scaffolding suffer a drop in performance and escalating input token cost due to the static weight of LLMs. This research underscores a fundamental premise of our work at Boltzbit to achieve General Learning Intelligence and democratise model ownership. Scaling up static-weight model size or layering on more workarounds cannot address the spiraling cost of training AI systems. The viable path is a new AI architecture that adapts its own weights. Preview version of the paper: arxiv.org/pdf/2609.18842

A1 is making live avatars hard to keep up with. Literally. My hiking buddy challenged me to a race and sprinted off down the trail without me. That’s a different league of character interaction. He spins all the way around to check the view behind us and jumps so hard he scares off a squirrel, while his water bottle stays the same bottle through every step, turn, and jump. All of this happens in real time as we talk. I told him to look at the clouds. He said they looked like giant marshmallows, then started panicking that it might rain. I can cut him off mid-sentence or change the plan whenever I want. Apparently, he can change the plan too. I thought we were hiking, not racing. Next I want a camping ghost that tells stories back. Think you can keep up with him? Meet him in the @VivixLabs_HQ Playground. 👇