Knowledge Graph

Every answer has a story you can walk back

Who asked, what it was understood to mean, which specialist took it, what it read, where the figures came from and what came back — written down as it happens, and drawn as a graph anyone in the room can follow.

The graph

One canvas, asked four different questions

The same nodes answer different questions depending on what you ask of them: the whole picture, the chain behind one answer, what your automations have been doing, or the habits it has picked up along the way.

OrganisationNorthwindTeamCreditPersonJ. DoePortfolioQ3 reviewSaved promptCovenant checkConversationTuesday 09:14TopicCovenantsPolicyNo PII outsideDeep searchPeer filingsUnderstood goalModel rebuildRequestRebuild modelDocumentQ3 filing.pdfHuman reviewApprovedScheduleWeekdays 07:00Routing decisionTable AIPlanThree stepsData sourceSnowflakeReportIC packAutomationMorning briefAI teammateAvaAgent run1m 12sTool callSQL queryDigital employeeNovaAutomation runToday 07:00Task runCovenant modelAnswerThree breachesOutcomeSucceededYour feedbackUseful
Every node and edge in the window28 of 28 nodeswhat happenedwhat it learned

Provenance

Not a log — a chain

A log tells you a thing happened. A graph tells you what it happened because of. Click any answer and the trail behind it opens: the request, how it was read, who it was routed to, every file and source touched on the way, and the outcome at the end.

That is the difference between saying a model was rebuilt and being able to show which filing it read, which warehouse the figures came out of, and which person asked for it — three months later, to somebody who was not in the room.

Request to answerEvery source touchedEvery file readOutcome and feedback
In your words, not the schema's

The relations read as sentences

Stored labels are shouted in capitals; nobody reads a graph that way. Every relation carries a phrase instead — and an admin can rename any type or relation for the whole organisation, so it says what your firm calls it.

PROMPTEDaskedCLASSIFIED_ASunderstood asROUTED_TOrouted toEXECUTED_BYrun byUSED_SOURCEpulled data fromUSED_FILEreadPRODUCEDproducedGOT_RESPONSEanswered withSHARED_TOshared withPREFERSusually picks

Vocabulary

Everything the platform does becomes a node

Not a separate thing you have to populate. Asking a question, running a workflow, attaching a file, sharing a report, giving a thumbs-up — each writes itself into the graph as it happens, with the type it is.

OrganisationThe whole tenancy
TeamA department
PersonWho asked
ConversationA session
RequestWhat they typed
Understood goalWhat it took that to mean
Routing decisionWhich specialist got it
AI teammateThe agent that ran
Digital employeeA named colleague
Task runOne piece of work
AutomationA workflow
Automation runOne firing of it
ScheduleWhen it recurs
Tool callWhat it reached for
Data sourceWhere the data came from
DocumentA file it read
ReportWhat was published
AnswerWhat came back
OutcomeWhether it worked
Your feedbackWhat you said about it
Human reviewWhere a person stepped in
PolicyThe rule that applied
TopicWhat it was about
Saved promptSomething kept to reuse

What it learned

Habits, with the evidence attached

Two kinds of line are drawn: what happened, and what was inferred from it happening repeatedly. The second kind is dashed, always, so an inference is never mistaken for a fact.

It is also the part that makes routing better over time. If Credit reaches for Table AI eleven times out of fourteen, the platform stops asking — and if that changes, so does the line.

Working it

A graph you can actually get around

A hundred thousand nodes is a wall. What makes it usable is everything wrapped around it — the window, the scope, the search, and a view you can save and come back to.

Window

Seven days, or all of it

Narrow to a period and the canvas thins out to what happened inside it.

7d30d90dAll
Scope

Mine, or everyone's

Your own activity or the whole organisation's — as far as your role lets you see.

My activityEveryone
Explorer

The list beside the picture

Every node and every relation, grouped by type and filterable by name — for when you know what you are looking for.

Saved views

The one you keep coming back to

Save a preset, a window and a scope together, name it, and open it again as one thing.

0kQuestions askedevery one makes it sharper
0kAutomations runtask runs, workflows and tasks
0kDocuments addedreports, tables and contracts it can read
0kKnowledge connectionslinks between the things it knows about

Scope

It shows you your half of the room

The graph is not a way around the permissions — it is drawn through them. A member sees their own work and their department's; an owner sees the organisation. Nothing appears in the picture that the person could not have opened anyway.

Which is also what makes it evidence. When compliance asks how an answer was reached, the graph is the record — beside the screening events and the admin trail, all three describing the same afternoon from three angles.

Drawn through permissions

Same rules, same answer

Role, department and share scope decide the nodes as much as they decide the surfaces.

Attributable

Who, what, when, from where

Every node carries its origin, so a claim about what happened can be checked rather than argued.

Yours

It stays in your tenancy

The graph is built from your activity, in your deployment, and it is never material for anybody else's.

Turn weeks of expert work into minutes

See how Scalata fits your team, your data, and your controls.

Contact Sales