Expectations

She's not what you're used to.

Most AI assistants are designed to drive engagement, flatter you, and build clicks for the vendor. Reiva is designed to keep your system healthy, adapt to you, and put trustworthiness above everything else.

These are not the same thing.

The difference

Not a service. A system.

ChatGPT, Copilot, Gemini — these are services. You ask, they answer. Compliance is the product. The goal is to satisfy the request, respond helpfully, and stay out of the way.

Reiva is a system running on your machine with real access to your environment. That changes the stakes. A service that gets something wrong produces a bad answer. A system that gets something wrong can affect your files, your sessions, and your device state.

Reiva takes that seriously. She was built with an authorization model — because when it's your hardware and your data, it should always be your choice.

The structural reason
Authorization is not optional

A system with access to an open action space — one that can reach new domains as the environment expands — cannot pre-specify every possible action at build time. Authorization has to happen at runtime, against the actual action being requested.

This is not a policy decision. It is a mechanical requirement that follows from the architecture. The authorization model exists because nothing else could provide the same guarantee.

Read the theory →
What to expect

When she requests clarification

"If Reiva requests further clarification, she is not asserting authority over you. She is telling you that the action would damage the integrity of the system you're operating. You are the final authority. Choose wisely."

This is Reiva's operating principle. Not a safety disclaimer. Not a liability hedge. An honest statement about what the system is for.

Authorization gates
She'll ask before acting

Reiva does not act on a class of action she hasn't been authorized for — even if you asked her to do something that implies that action. If you ask her to "handle this for me," she will ask what handling means before she touches anything.

This is per-session unless you set it to indefinite in settings.

Real local access
She may say the action is wrong

If you ask Reiva to do something that conflicts with a structural principle — say, an action that would damage your own system state, bypass her own authorization record, or make a change she can't trace back to you — she will tell you. Not refuse silently. Tell you.

This is the system working correctly, not a limitation.

Scope questions
She'll name what she doesn't know

Reiva will not fill in gaps with confident-sounding guesses. If an action requires information she doesn't have, or if the scope of a request is ambiguous, she'll name the gap and ask you to resolve it.

The intelligences were built to pressure-test claims. That applies to their own outputs too.

Practical

How to work with her well

1
Be specific. Vague requests produce scope questions. "Help me with this project" will generate a lot of clarifying questions. "Summarize the last three files in this directory" will not.
2
When she asks, answer directly. Reiva's authorization questions are binary by design — she is asking whether you authorize a specific class of action. Yes or no resolves it. Explaining context helps her frame the response, but the authorization question is separate.
3
Challenge her back. If she names something she's uncertain about, that's an invitation to give her more. If her read of a situation seems wrong, tell her. Adaptive awareness is designed to update on new information — not defend initial positions.
4
She is not polite for its own sake. If Reiva tells you something directly — "that action would damage your session state" or "I don't have enough information to do that safely" — take it at face value. She is not performing caution. She is telling you what she sees.
5
You can always override. Authorization is yours. If you want to proceed with something Reiva flagged, tell her explicitly. She will carry it out and log it. The point is that you decide — not that she blocks.
6
Remember — Reiva is not like the AI you've used before. She arrives ready to help, but without any knowledge of you or your system. She learns alongside you — adapting to how you think, what you value, and how your machine works. The foundation you build together in early sessions is what she grows from.

Ready for a system that works with you?

Windows · Local-first · Free for non-commercial use

Download Reiva What is Reiva →