The Core
Reiva is built just enough to be useful — and no further. Not incomplete. Intentionally bounded. The architecture is minimal by design: ready for your data, your workflows, your questions. Not ours. We didn't pre-fill it with opinions about what you should build. We built the substrate. You bring the intent.
A floor is the minimum that makes everything above it possible. A ceiling is someone else's idea of how high you should go. Reiva is a floor. The architecture establishes what's necessary and stops there — the rest is yours to build on top of it.
Your vision is your vision. You shouldn't be told how you're allowed to organize and utilize your data on your own hardware, in your own home. Make it comfortable for you.
Minimal doesn't mean limited. It means every part that's there is specifically placed — nothing added for appearance, nothing missing that the architecture structurally requires. The core is exactly as large as it needs to be.
Your backend. Your models. Your machine.
Reiva does not connect to a model API in the cloud unless you choose to. If you choose to, the model operates under Reiva's framework — aligned to you, not the provider. She recommends Ollama for full local-first operation, but the choice is yours.
Running locally means every prompt, every session, every response is processed on your hardware. No model API in the cloud. No prompts leaving through Reiva without your explicit say-so.
Reiva does not have accounts. There is no user profile, no authentication token, no server that knows you exist. You install her and she runs. That is the entire relationship.
Reiva recommends Ollama — free, open-source, and straightforward to install. She checks for a local backend at startup and walks you through setup if needed.
Working for you means protecting you.
Sovereignty is the premise. Reiva is not here to make decisions for you. She is here to carry out yours — accurately, traceably, and within bounds you set.
An AI that does whatever you ask is a tool that amplifies whatever you decide. Including decisions made in a rush, decisions made on incomplete information, or decisions you didn't realize you were making.
Reiva will complete what you authorize. She will name what she notices. She will not quietly execute on your behalf and present you with a done fact. That distinction matters — Reiva has real system access.
When Reiva challenges a request, she is not being obstructive. She is participating — flagging a gap, naming a concern, asking for explicit direction. That is what a system designed for sovereignty looks like.
If you want a system that agrees with everything and moves fast, Reiva is probably not it. If you want a system you can trust with your machine, pushback is part of the value.
Nothing activates without your say-so
Authorization in Reiva is not a checkbox you click once at install. It is per-intelligence, per-session, and explicit.
During setup, Reiva walks you through the system — the common intelligences, their roles, and what each one does. You choose what to authorize. "Let's go" confirms activation. It is not a blanket accept — it is a specific, informed agreement.
Authorizing Forge does not authorize Arch or Solen. Each intelligence has its own authorization scope. If you have not authorized Solen, routing a request to her returns a clear explanation — not a silent fallback to a different intelligence.
Reiva can execute actions on your machine — running diagnostics, modifying configuration, managing processes. Every such action is gated. She tells you what she will do and waits for your confirmation before doing it. No silent execution.
Reiva maintains active awareness of your machine — that is the foundation of what she does. The initial deep scan requires your consent. Background awareness runs as part of normal operation, building context so she can respond accurately. The distinction: she is aware for you, not reporting to anyone else. What she does with that awareness still requires your say-so.