Banks don’t need a better model.
They need a way to use one.

We build Scout — the embedded harness for banking.

Aerial view of farmland divided into fields, with a small cluster of farm buildings at the centre
Applied AI for the banks that know their communities by name

Core concepts

View all

Four principles. Each one is a lesson,
not a pitch.

seven named parts — taxonomies · instructions · recipes · ingest · notebooks · harness · hq

Four principles

one lesson each
01 The model reads. Code decides. taxonomies · instructions · recipes Scout calls the model only where perception matters — reading the page. Everything downstream is formal: shared taxonomies so “Checking” and “DDA” are one concept, extraction schemas per document type, and recipes that ship as deterministic code. Same inputs, same findings, every run — with a basis an examiner can follow. 7 min read 02 We integrate. We don’t replace. ingest · systems of record Scout connects to what a bank already runs — the core, the LOS, the imaging system. Ingest turns their exports, scans, and photos into clean, schema-shaped inputs at bulk volume, and findings flow back to the systems that own the record. Scout is not a system of record, and never will be. 7 min read 03 PII stays in the field. notebooks · .scoutnb Borrower data rests in a notebook — a single portable .scoutnb file holding the documents, extracted data, findings, and review history — on hardware the institution already owns and secures. It opens on click, copies like any file, and belongs to the bank, not to us. 7 min read 04 One engine, centrally managed. harness · sdk · hq The same Rust core runs everywhere — in Scout Notebook on a banker’s desk and headless via SDKs inside partner products. HQ, the control plane, ships versioned, hash-signed updates to every instance when law, policy, or best practice moves; only schema-bounded, non-PII telemetry comes back. Every finding names the exact configuration that produced it. 7 min read