Compare
Where Selira sits, and what it is not.
Selira overlaps with several categories and replaces few of them. Here is the honest version, by category rather than by vendor.
| Category | What it is genuinely for | Where it stops |
|---|---|---|
| Data platform / lakehouse | Holding enormous volumes cheaply and running analytics over them. Built for throughput and cost per terabyte. | A table can show you what it said last March. It cannot tell you what was true in March as you know it now, whose word each row is, or that rows from two systems are one thing — unless your team builds and maintains that, table by table. |
| Knowledge graph | Modelling arbitrary relationships and reasoning over them. Closest category to us conceptually, and the standards work is excellent. | A graph can carry time and sources; whether it does is a modelling choice made property by property, and asking it as of a date, under a ranking, is left to every query writer. |
| ERP | Running the business. Transactions, compliance, finance, procurement — decades of accumulated correctness you should not try to rebuild. | It keeps its own history well — change documents, posting and entry dates — but it is one system's view. It cannot represent another system disagreeing with it, because it was built to be the answer. |
| Master data management | Matching records across an estate, deciding which one wins, with a steward for the hard cases — and making that decision stick. | It decides once, for everyone, and the survivorship rule becomes the answer. Selira decides per question: a match is a claim with its confidence, a person's confirmation outranks a score, and nothing is merged away. |
| Digital twin | Modelling an asset or a network closely — its physics, its data, its topology and traffic — and simulating how it behaves. Genuine engineering, and not what we do. | Often a model of one asset or one system, fed by its own telemetry. It rarely models the operation around it, whose word a reading is, or a disagreement. |
| Selira | Answering one question across time, space, truth and source — the records your systems hold, taken in with their origin, and the context around them, in one model. | Not a warehouse, not a physics simulator and not a planning solver. It cannot know what nobody recorded anywhere. |
The actual difference
Most dimensions are modelled well somewhere. All four on one statement is rare.
Every system in the list above is excellent at its job. Most operational systems store a fact as a record with one current value; its history, its provenance and its source sit beside it, where they are kept at all. Where they are kept, it is as conventions — built table by table, done differently by each team.
That is why crossing time, space, truth and source is a project every time: the conventions do not compose. In Selira they are the statement model itself, with one query path over all four. An AI agent put on top of your systems inherits the same split: it can read the records and the documents, but not whose word each is or when it held.
We are not trying to be a better version of any of these. We are a model that can answer across them — and hold the context none of them has a place for.
When something simpler will do
Modelling an operation has a price. Where a simpler tool answers the question, it is the better buy.
One system, one truth
If your operation runs on one system and nobody disputes its numbers, a model around it costs more than it returns. Use the system you have.
Analytics at volume
If the question is aggregate reporting over billions of rows with no provenance requirement, a warehouse is the right tool and will be cheaper.
Physical simulation
If you need to know how a component behaves under load, you need engineering simulation. We model the operation around it, not the physics inside it.
Test it against what you already run.
Bring a question your current stack cannot answer in one go. If it turns out your stack can, we will tell you.