How does it run entirely locally? Don't LLMs need massive hardware?
+
Modern quantized models run at impressive capability on consumer hardware. Semblance uses llama.cpp on desktop (Windows, macOS, Linux) and MLX on Apple Silicon, routing tasks to the right model for the job — fast 1B models for classification and triage, larger models for reasoning and drafting. On Apple Silicon Macs and modern PCs with 16GB+ RAM, performance is excellent. We publish detailed hardware profiles so you know what to expect before you commit.
How do I know it's actually not sending my data anywhere?
+
The core is open source. Every network connection Semblance is permitted to make is declared in the Gateway — the only process with network access. The AI inference core is architecturally isolated at the OS level. You can verify this in the code, and the privacy community will. We encourage it. The repository also includes automated privacy audits that run on every build and can be run locally.
What happens to my data if I stop using Semblance?
+
It stays on your device, because it was only ever on your device. There is nothing to delete from a server because nothing was ever sent to one. You can export your complete knowledge graph at any time, or delete it entirely. We have no copy. We've never had a copy.
What's the difference between Guardian, Partner, and Alter Ego modes?
+
Guardian mode: Semblance drafts and surfaces everything, approves nothing autonomously. You're in control of every action. Partner mode: Semblance acts autonomously on low-stakes, routine tasks (subscription cancellations, scheduling, follow-ups) and surfaces high-stakes actions for your approval. Alter Ego mode: full autonomous operation — Semblance runs your inbox, calendar, and correspondence, reporting back in the Morning Brief. Every mode maintains a complete audit trail. You can switch tiers at any time, per-domain.
Is the Founding Member offer really limited to 500 people?
+
Yes, and the cap is honest, not manufactured urgency. 500 founding members generating real, authentic first reactions is how this product launches. It's not "only 12 left" countdown-timer dark-pattern scarcity — Semblance's own Adversarial Self-Defense feature would flag that. The cap is real. When it's full, it's full. The $199 lifetime tier converts to the standard $349 lifetime after launch.
When does it ship?
+
Semblance is feature-complete. All 33 steps are done with over 3,800 passing tests. We're in final polish, hardening, and pre-launch preparation — not building, fixing. The target is a non-technical user downloads it, it runs in five minutes, and they never go back to ChatGPT. Join the waitlist and you'll be first through the door.
What platforms does it support?
+
Desktop: macOS, Windows, and Linux. Mobile: iOS and Android as a companion to your desktop (with on-device inference for lightweight tasks on capable hardware — full 3B models on Apple Silicon iPhones). The desktop and mobile apps are peers, not main-and-companion — both can run inference and both maintain a synchronized local knowledge graph.
Who built this?
+
Semblance is built by Skyler Kruger, founder of VERIDIAN SYNTHETICS — a one-person operation in Bend, Oregon. No formal technical background. No VC. No institutional backing. Just a conviction that the AI assistant category is being built wrong, and the determination to build it right. The founding story is part of why this product exists: access to the kind of AI that actually manages your life shouldn't require a subscription to a surveillance company.