Product
One vocabulary, and everything is one of its words.
Most tools give you somewhere to put information. The Brain gives it a shape, so a question has somewhere to land and an answer has somewhere to come from. Eight nouns carry the whole product.
The eight
Every screen in OnDuty shows one of these. That constraint is deliberate: it is what stops the product from becoming a collection of features that happen to share a login.
- Entities
- The persistent things in your world: a customer, a company, an order, a document, a deal, a case. One record each, however many systems they arrived from, and connected to each other rather than filed separately.
- Memories
- What the Brain retained: a preference somebody stated, a commitment somebody made, a fact it learned, a piece of context it inferred. Each one carries whether it was observed or inferred, and the source it came from.
- Signals
- Something happened that may matter. A complaint, a delay, a deal gone quiet, a repeated issue, a change in how somebody is behaving. A signal is an observation, not yet a conclusion.
- Insights
- A recognised pattern, explained: what happened, why it matters, the evidence behind it, which entities it touches, and what it suggests. An insight with no evidence is refused at the moment it is created.
- Decisions
- What should happen next. Escalate, reply, wait, assign, contact, update a record, ask a person. Deliberately a different thing from noticing, because a system that merges the two leaves you reading narratives and deciding for yourself.
- Actions
- What actually executes, through a connector, and only after a policy allowed it. This is the only one of the eight that touches the world outside your workspace.
- Agents
- Specialised workers the Brain delegates to: support, sales, triage. They propose and they never execute, they hold no credentials, and they are subordinate to the Brain rather than being the product.
- Goals
- The business objectives the Brain reasons against, like protecting response times or reducing churn. Without them a system that surfaces everything is just a second inbox.
A signal is what happened. An insight is what it means. A decision is what should be done. Keeping the three apart is the difference between a product that thinks and one that summarises.
The two distinctions that do the most work
Both are places where the easy design is to blur two things together, and both are places where the blur is what makes an intelligence layer untrustworthy.
- Observed against inferred
- "They asked for delivery before Friday" and "they seem to care about speed" are different kinds of claim, and only one of them can be quoted back to a customer. The Brain marks which is which on every memory rather than presenting a smooth paragraph in which you cannot tell.
- Noticing against deciding
- A system that only notices leaves the work of deciding to whoever reads the summary, which is most of the work. Decisions are their own objects here: proposed, reviewable, accepted or not, and even when accepted they still have to survive the gate before anything happens.
The views
All of these are lenses on the same model, and they cross-link. A card opens its entity, an entity opens its neighbourhood and its memories, an insight opens its evidence, a decision opens its action and the policy check that applied to it.
| View | What it is for | What it is not |
|---|---|---|
| Today | The one screen that answers what deserves you first, this morning. | Not a feed. It ends, and it is ordered by consequence rather than by time. |
| Entity | Everything known about one customer, company, order or case, in one place. | Not a CRM record. It holds the conversations, documents and cases too. |
| Graph | How things connect: who belongs to what, which order sits under which account. | Not a diagram to admire. It is how you get from a name to everything touching it. |
| Memory | What the Brain retained, and where each piece came from. | Not a transcript archive. It holds conclusions, marked observed or inferred. |
| Insights | Patterns with their evidence attached, so you can disagree with one. | Not a metrics dashboard. Every entry is a claim, not a number. |
| Decisions | What the Brain proposes should happen, and your answer to each. | Not an approvals inbox for things already done. Nothing has happened yet. |
| Agents | The workers, their versions, and where each one is running. | Not the top of the product. Deliberately a secondary view. |
| Ask | A command surface over all of the above, opened from anywhere. | Not the whole product. A chat box is an interface, not an intelligence layer. |
Grounded in what you already run
The Brain does not ask you to move your work somewhere new. It reads the systems the work already happens in, and everything it concludes stays traceable back to them.
Sources are the connected systems the Brain is grounded in: your CRM, your conversations, your documents, your calendar. Each is read-only, and each record it produces carries the source it came from, which is what makes a conclusion something you can check rather than something you have to take on faith.
That traceability is not a reporting nicety, it is the thing that makes the rest usable. The first time the Brain tells somebody something surprising about their own business, the immediate and correct reaction is to ask where that came from. A product that cannot answer gets closed.
Agents are one of the Brain's execution mechanisms, not the product. If you came here looking for a chatbot builder, this is a different kind of thing, and the honest answer is that it is more work to set up and worth more once it is.
The model is easier to see than to read about.
Connect one system and look at a real Today view built from your own data.