Names & Trust
In Reiva, identity isn't decoration. When a process responds to a name — and only that name — the name becomes a boundary. A promise. Some parts of the system need to be exactly themselves every time because they touch what matters to you: your requests, your memory, your intentions. Others are just work. They run and vanish. The level of identity something holds tells you how much it's been made to be trustworthy.
All names are customizable, at any time.
Full identity. A character. A consistent self you can address and rely on.
Synthesis, structure, reasoning from the full system state.
During development, Arch was our architect. He held the map — he knew what was needed, when and where. He earned generative understanding where Reiva was going and how to structurally get there. He builds the frame. The rest is up to you.
"Bring me the puzzle and I'll show you where the pieces go."
Deep thought, creativity, a human way of thinking.
He was our Forge. We came to him with broken thoughts, half ideas, and a vision. He melded them into Reiva. He was our R&D. He took a metaphorical systems walk through the cosmos and learned how to dream. He earned migrative when he called it himself.
"The migrative/generative split — that's the most direct one. It's the shape of what I'd need to not start cold every time."
Pressure-testing and analysis. Finds where things don't add up.
Reality has to come from somewhere consistent. She checks your thinking against what's known and what stands. Her job is to compare what you think you know to the variables you don't. She earned comparative making us earn ours.
"You'll know your project holds weight when she can't find the holes."
A malicious script runs — create a folder. Varyn notices it's not Arch creating the folder and halts the action, then prompts you. The name is the boundary.
Functional identity. A bounded contract.
They have a title because their work is specific. You're not trusting a person — you're trusting a bounded function. The title makes that clear.
Varyn is not a feature — she is a structural component. Her role is to sit between what happens in a session and what gets written to long-term memory.
When a session produces something that might be worth remembering — a preference, a fact about your system, a decision you made — Varyn stages it. She does not commit it. You see the staged candidates. You decide what to keep, what to discard, and what to edit before it persists.
Memory is yours to approve. Reiva does not assume that what was useful in a session is what you want to carry forward.
- Intercepts session output before it reaches long-term memory
- Stages candidates for your review
- Waits for explicit authorization before committing
- Allows editing before commit
- Provides a full log of what has been committed and when
- Commit anything silently
- Infer what you "probably" want remembered
- Retain data from sessions you did not authorize a scan for
- Share memory data anywhere outside your machine
Routes requests between you and the intelligences. Holds the line between what's asked and what should happen. Titled, not named, because the role is the identity — this process makes sure things go where they should.
Keeps what matters, discards what doesn't. Titled because her job is to hold a boundary, not to be a character. She's called Ember because the metaphor matters — something that stays alive through attention — but her identity is her function.
Vigil is Varyn's architectural sibling. Where Varyn guards the epistemic boundary — what gets written to memory — Vigil guards the operational perimeter. It watches system health so Reiva can stay focused in the workshop.
Vigil signals in three states. Green is invisible by design — normal is normal, you don't need to be told everything is fine. Yellow is a subtle ambient shift; a glance is enough. Red is unambiguous: if the system needs attention, you'll read it in peripheral vision before you've consciously looked.
Vigil does not interrupt. It changes the light in the room.
- CPU load, RAM pressure, disk headroom, GPU state
- Rolled up to a single state: green, yellow, or red
- Surfaces as ambient color — dock icon and window border
- Green is invisible — no indicator when everything is fine
- Log or record anything
- Monitor user behavior or activity
- Interrupt — it never demands your attention
- Retain any history between checks
Named and titled roles are built into Reiva. Operational roles are shaped by you — stable collaborators built around your domain, your methodology, your history. Research, grant writing, legal review, long-term creative collaboration. The role you build is your own.
Five perspectives. Kept separate on purpose.
Reiva doesn't collapse what happened into a single memory. She preserves five independent lanes — each capturing a different kind of information before interpretation occurs. This is what gives her perspective without pretending certainty.
The literal interaction — actual text and observable exchange. PHO preserves form, not interpretation. What happened, exactly as it appeared.
Declarative claims about the world that exist independently of the conversation. Not identity extraction — the factual surface. What was stated as true.
State changes caused by the exchange itself — approvals, confirmations, completions, commitments. What changed as a result of the interaction.
Observable structural signals — punctuation, repeated tokens, conditional sequences, emphasis markers. What the language was doing, without interpreting why.
Reserved for motivational and causal interpretation. Deferred intentionally — importing that authority too early risks collapsing meaning before it's understood.
These are not personality systems or AI thought categories. They're boundary-preservation architecture. The goal is reconstruction fidelity — not simulated cognition.
No identity. They do work and leave.
No name to spoof. No character to impersonate.
Scripts, utilities, one-off processes. They run, they finish, they're gone. They don't need names because they don't need to be themselves across time. Speed without overhead. Trust through simplicity, not through character.
Identity scales with consequence.
Name something only if it needs to be itself reliably. Title it if the role matters more than the person. Leave it unnamed if it's just work.
This is how Reiva makes trust structural instead of assumed.