A sovereign assistant for knowledge workers — on hardware your business owns or leases.
Voice-mode chat, persistent knowledge library, projects, skills, image generation. Tied into your calendar, your tasks, your inbox. The whole thing runs on RadiantCore infrastructure in St George — your data never crosses a boundary you didn't authorize.
The assistant, the library, and the workflows — on one stack.
Real full-duplex voice. Local STT, local TTS, streaming responses. Hands-free in the truck, the office, or the cockpit.
Codex of every fact, decision, and document the assistant should know about. Indexed for retrieval, owned by you, never shared across customers.
Threaded conversations, scoped to a project, with a task system the assistant can read and write. Pulls in your calendar and email when you ask it to.
Reusable prompted workflows the team builds together. "/morning-briefing" pulls calendar + tasks; "/preparing-meetings" pulls attendee correspondence. Your skills, your library.
FLUX on dedicated GPUs. Generated images stay on the rack; no third-party prompt logging, no aggregated training set.
Record a meeting; the assistant transcribes, identifies speakers, summarizes, and surfaces action items. The audio + transcript stay on your hardware.
Local LLMs by default. External APIs only when you say so.
The default model is a local LLM running on RadiantCore GPUs. Your prompts never leave the rack. If you want an external model — Anthropic, OpenAI, anything else — that's an admin-gated escalation per project, with a clear UI affordance and an audit trail. No silent egress.
TTS and STT are local. Web search routes through a self-hosted SearXNG. Calendar and email integrations use OAuth tokens you authorize per user. Every external surface is a deliberate choice with a flag you can flip.
How you get a Radiant Core instance.
We provision your tenant on our hardware, set up users + admin, configure the integrations your team uses, and pre-seed the codex with your domain knowledge.
Every customer gets isolated tenancy at the database level. Your codex never surfaces in another customer's retrieval. Your conversations never train someone else's model.
We monitor, patch, upgrade. New features land for everyone. Your skills + codex are yours; the platform improvements are ours and they keep coming.