Documents in. Canonical, versioned data out.
We do not produce financial information. We take what authoritative institutions publish and turn it into a representation software can use — while keeping the original as evidence.
Publications from authoritative institutions are captured as they are issued, with their date and origin recorded.
Structured facts are extracted from documents. The document itself is retained — it is the evidence, not a by-product.
Facts are mapped onto one canonical model: one record per instrument, one per issuer, identifiers reconciled across sources.
Rules, cross-source checks and human review run before anything is published. Unresolved conflicts are flagged, not guessed.
Data is versioned and delivered through APIs and files, with every field still pointing back to its source.
Why an institution could rely on this.
Trusting a third party with data that feeds regulatory and client work is a serious decision. These are the properties we think earn it — and the ones we want you to hold us to.
- Traceability
Every value carries its origin. “Where did this come from?” resolves to a specific document and passage, not a person’s recollection.
- Versioning
Nothing is overwritten. The state of a record on any past date remains queryable, which is what reconstructing a report requires.
- Explicit uncertainty
Where sources disagree or a document is ambiguous, the platform says so rather than picking silently.
- Boundaries
We normalize and distribute. We do not interpret, value, rate or advise — those judgments stay with you.
Standing on established practice.
Modelling financial reference data is a solved problem elsewhere. We study how the established data infrastructures and ontologies do it, and follow their conventions where they fit the Tunisian market rather than inventing our own.
We are building this with institutions, not for them.
Tell us what reference data costs your team today — the hours, the breaks, the workarounds nobody has time to fix. That is what decides which domains we build first.