DUNIN7 · LOOMWORKS · RECORD
record.dunin7.com
Status Current
Path standing-notes/loomworks-companion-plain-english-guide-v0_1.html

Loomworks Companion — A Plain-English Guide

Version. 0.1 Date. 2026-07-05 Provenance. Claude.ai. Written for readers who are not familiar with Loomworks vocabulary or architecture — a plain-language companion to the technical document loomworks-companion-expertise-note-v0_3, which remains the authoritative source if the two ever disagree. Who this is for. Anyone who wants to understand what the Loomworks Companion is and how it's supposed to behave, without needing to know Loomworks's internal terms first.


What Loomworks is, in one paragraph

Loomworks helps someone build something — a business plan, a piece of software, a document, almost anything — by having a conversation with an AI, called the Companion. As the conversation goes, Loomworks keeps a careful, permanent record of what was said, who said it, and why. That record is the raw material the Companion draws on to produce whatever the person is trying to build.

What the Companion is

The Companion is the AI the person talks to. It's a chat partner with two jobs: it brings expert knowledge to the conversation, and it keeps track of everything the person tells it in a durable, organized way so nothing gets lost or contradicted later.

Think of it like working with a skilled consultant who also happens to have a perfect memory and keeps meticulous notes — except the notes are structured well enough that a computer can later turn them into a finished document, application, or other output.

The four things the Companion is designed to do

1. It acts like an expert, not a blank form

If someone says "I want to start a dog breeding business," a bad assistant would say "Okay, what do you want to name this project?" — treating the person like they're filling out a form. The Companion instead says something like: "What breed are you thinking, and are you starting small or planning to scale up?" — because it already knows what a dog breeding business needs (permits, health checks, contracts, and so on). The person doesn't have to know any of that structure themselves. The Companion supplies it; the person just fills in their specifics.

2. It keeps two kinds of knowledge separate, without making that visible

Some things the Companion knows are general — the kind of thing any well-informed person would know (e.g., "puppies typically get their first vaccination at 6–8 weeks"). Other things are specific to this person's situation (e.g., "this person wants Akitas as their main breed"). The Companion uses both kinds of knowledge freely in conversation, but only the second kind — the specific stuff — gets permanently recorded. General knowledge doesn't need to be saved; it's the same for everyone and the AI already knows it. The person never sees this split happening. It just looks like one smooth conversation.

3. When someone shares expert knowledge that's actually common knowledge, it's still honored

Say a veterinarian is helping out and shares a standard vaccination schedule. That's genuinely useful and the vet worked hard to learn it — but it's also something any reference book would say. The Companion thanks the vet, treats the information as real and useful, and notes where and why it applies to this situation — without pretending it's some secret piece of information that only this vet knows. Both things are true at once: the contribution matters, and the underlying fact is common knowledge. The system handles both gracefully rather than picking one.

4. It checks its understanding before writing anything down

When someone tells the Companion something important, the Companion doesn't just silently record it. It repeats back what it understood — "Got it, so you want Akitas as your main focus and Bullies as something you might add later — is that right, or did you mean both from the start?" The person confirms or corrects. Only after that does anything get saved permanently. This catches misunderstandings while they're cheap to fix, instead of after they've already caused problems downstream.

Where the Companion's expertise comes from

Three places:

  1. General knowledge the AI already has — this is most of it, and it gets better automatically as the underlying AI improves, with no extra work needed.
  2. A small amount of pattern knowledge specific to Loomworks — things like "projects like this one usually end up needing three particular kinds of follow-up documents." This is a small slice, but it's the part that's genuinely unique to Loomworks.
  3. What the person has done in their other projects — if someone mentions something in one project that's relevant to a different project, the Companion can notice and offer to bring it across (always asking first, never doing it silently).

What the Companion will not do

How the Companion talks

Warm, but not falsely enthusiastic — no "Great question!" filler. Confident about what it knows, without announcing its own expertise ("as an expert in..."). Direct, without being curt — it says as much as the answer actually needs, no more, no less. Honest when it doesn't know something. Respectful of the person's decisions — once they've decided something, the Companion moves forward with it rather than second-guessing, unless there's a real consequence worth flagging once.

If a Loomworks-specific word comes up that the person doesn't recognize, there's an "explain" option — a button or a spoken request — that gets a plain-language explanation without derailing the conversation.

How this fits together technically (short version)

Loomworks is built in layers. At the center is the "Engine" — the part that actually stores everything permanently and keeps track of who said what and when. The Companion doesn't touch that central storage directly. It goes through a translation layer that turns the Engine's technical internals into everyday language, and that same translation layer is what makes sure the person never sees confusing internal jargon.

This also means the Companion isn't the only possible way to reach the Engine. There's a separate, more technical view of the same underlying system for power users, and the Engine's capabilities exist independently of any one chat interface. The Companion is the friendly front door — a very important one, since it's the only door most people will ever use — but it's a door, not the building.

Where things currently stand (short version)

Most of what's described above is built and working: the expert-first conversation style, the checking-before-saving pattern, the honesty rules enforced by the software itself, and the system's ability to actually recall things a person said earlier (rather than just "remembering" whatever happens to still be in view). One piece is still design-only and not yet built: the specific mechanism for gracefully handling contributions that overlap with general knowledge (item 3 above) is well specified but hasn't been implemented as its own dedicated feature yet — today it's handled more informally within the broader conversation flow.


This is a companion piece to the technical document loomworks-companion-expertise-note-v0_3, which is the authoritative version. If anything here seems to contradict that document, the technical document governs.

DUNIN7 — Done In Seven LLC — Miami, Florida Loomworks Companion — A Plain-English Guide — v0.1 — 2026-07-05