the primitive · tajuThe firm's own institutional brain, compiled.
Taju (Finnish: grasp, comprehension) is what Niska builds from a firm's own documents, conventions, and corrections — compiled, citation-anchored, and regenerated as the work evolves. It is never a system of record, and every claim inside it traces back to a source document a partner can check.
Taju is the firm's own grasp of its own work.
Taju is Finnish for grasp, or comprehension — not a stored answer, but an understanding a firm holds. That is the model. Taju is the compiled, citation-anchored institutional brain Niska builds from a firm's own documents, conventions, and corrections. It sits at the core of every Niska deployment: the engine that reads the firm's documents, and the thing everything else is built from.
It is regenerated, not authored — rebuilt as the firm's work evolves, so it is never a fixed export of what the firm once knew. Every claim inside it traces back to a source document a partner can check.
Built from the work, not around it.
Taju is not configured once and left alone. It is compiled from three things the firm already has, and rebuilt as they change. Nothing generic ever ships — what one firm's Taju knows belongs to that firm alone.
- 01
The firm's documents
Every email, contract, workpaper, and record the firm holds, read in full, inside the firm's own environment.
- 02
The firm's conventions
The way this specific firm does the work — its house style, its shortcuts, its judgment calls — learned from the work itself, not configured by hand.
- 03
The firm's corrections
Every time a partner fixes what Taju gets wrong, that correction compiles back in. Taju gets sharper for this firm specifically, not for firms in general.
Just as binding: what it is not.
- 01
Not a system of record.
Taju reasons over the firm's existing documents. It does not become the place they live, and it does not replace what a firm's actual systems of record hold.
- 02
Not a chatbot bolted onto storage.
Taju is compiled from the work itself — not a conversational front end wrapped around a file store.
- 03
Not authored content.
Nobody hand-writes what Taju knows. It is regenerated from the firm's own documents, conventions, and corrections.
- 04
Never answers without a source.
Every claim Taju makes traces back to the exact document and page it came from.
- 05
Never pools between firms.
What one firm's Taju learns stays inside that firm's deployment. Nothing generic ships, and nothing learned in one firm's environment reaches another's.
The firm's own memory compounds.
Generic AI resets with every session. Taju doesn't. Every correction a partner makes and every convention the firm has built over years compiles back into Taju, so the firm's institutional memory gets deeper the longer it runs — instead of starting over each time.
And because every answer traces to a page a partner can check, that memory stays accountable rather than becoming something the firm has to trust blindly. The firm's own conventions and corrections are the asset. Taju just keeps them answerable.
See what Taju would compile from your firm's own files.
Write to us, and we'll walk through what Taju would build from your firm's own documents, conventions, and corrections.