Creation
Reiva runs three intelligences on your local hardware, routes your requests through a structured pipeline, and gates every action — including memory — behind explicit user authorization. Here is the full picture.
Arch, Forge, and Solen
Each has a distinct role in the system. None of them activate without your explicit consent. Each is independently authorized — and each connects to the model you choose.
Arch is the generative intelligence. He synthesizes, structures, and reasons from the full system state. When you need to understand what is happening across your machine — not just a single process, but a pattern — Arch maps it.
He builds understanding from observation. He does not speculate beyond what the system state supports. His output is structured reasoning, not confident guessing.
When developing generative intelligence, we realized humans don't think that way. We don't arrive with clean, structured questions — we arrive associative, half-formed, somewhere in the middle of a thought. That's where migrative came from.
Your Forge takes raw data — a direction, a half-formed idea, a problem without a shape yet — and works it. You don't need to arrive with the answer, just the material. He builds the rest.
Solen is the comparative intelligence — she pressure-tests. When you have a plan, a diagnosis, or an output from Arch or Forge, Solen checks what holds against what is established and finds where things do not add up.
She is not an adversary. She is a second view with a different lens. Useful when accuracy matters more than speed, or when you want the uncomfortable question asked before you act.
Let's say you want to build something.
The intelligences each do their specific work. You decide the routing. You decide when it's done. The loop runs until you close it.
Receives the raw input. Builds structure from it — maps the space, organizes the data, lays it out so depth work can happen. Arch doesn't deep-think it. Arch makes it ready for the one who does.
Takes the structure Arch laid out and goes deep. Explores the space, pushes the ideas, finds what's interesting, builds from what's there. Forge doesn't just summarize — he advances.
Receives what Forge built and stress-tests it. Does it hold? Where are the leaks? What doesn't add up against what's established? Solen doesn't build — she finds where the build fails.
Takes Solen's findings. Cleans up what didn't hold. Patches the leaks. Sharpens what did hold. The work advances — it doesn't restart.
Run it again. The cleaned version — does it hold now? Are there leaks in what Forge patched?
It held. Take what survived and build the real structure around it. Arch turns it into something usable.
Each intelligence runs on the model you choose.
Forge, Arch, and Solen do not have to share a backend. Each pipeline connects independently — you decide what runs where.
Route Forge to Claude for continuity work, Arch to GPT for structural reasoning, Solen to a local model for pressure-testing. The pipelines are independent — the choice is yours.
Run everything locally via Ollama, or connect individual intelligences to external models. Reiva recommends Ollama for full local-first operation, but does not require it.
Model connections follow the same authorization model as everything else in Reiva. Each pipeline connection is explicit. Nothing connects silently.
Request to response
A request moves through Reiva's pipeline in defined steps. Here is what happens between input and output.
Input received
You issue a request. Natural language. No special syntax required for standard requests. Routing hints ("ask arch", "ask forge", "ask solen") are recognized if present.
Authorization check
The pipeline checks whether the target intelligence is authorized for this session. If not, the request is returned with a clear explanation. No silent rerouting.
Intelligence routing
The request is routed to Forge, Arch, or Solen based on explicit routing hints or the nature of the request. The Conductor determines which intelligence handles it.
Execution gate (if applicable)
If the response involves an action on your machine, the execution gate activates. Reiva describes exactly what she plans to do and waits for your explicit confirmation before proceeding.
Response and Varyn staging
The intelligence returns its response. If anything from this session is a candidate for long-term memory, Varyn stages it for your review. Nothing commits to memory without your authorization.
Ready to run her?
Windows 10/11 · Recommends Ollama · Free · Personal and research use
🐧 Linux users already know what to do. 🙂