A deterministic financial platform with an AI intelligence layer.

Upload brokerage and retirement statements, W-2s, 1099-DIV and 1099-INT forms, tax returns, trusts and wills. Files go directly to encrypted storage and are processed asynchronously.

Clariose validates each file, extracts native PDF text (with OCR fallback for scans), classifies the document, and runs a dedicated extraction schema for that document type — section by section for long documents.

Every extracted value keeps the page it came from and a verbatim excerpt. Clariose computes its own confidence score from text quality, schema validity, whether the value literally appears on the cited page, and cross-field checks — it never relies on the model’s self-reported confidence alone.

Extracted facts are normalized into one canonical model: people, institutions, accounts and owners, balances over time, securities and holdings, income, tax returns and forms, trusts, wills and beneficiary designations.

Entity resolution links the same person, account or trust across documents deterministically. Ambiguous matches — “Is John A. Smith the same person as John Smith?” — go to a review queue with the evidence side by side.

Every record shows where it came from: which documents, which pages, and when it was last observed.

Deterministic rules compare information across sources: statement dividends vs. Form 1099-DIV, statement interest vs. Form 1099-INT, balances reported by different sources, expected tax forms, and beneficiary coverage.

Tolerances are explicit rule configuration, and every result records the rule, its version, the values compared and the difference. AI is never used for arithmetic.

Reconciliation results become an action-oriented findings inbox, ranked by severity and financial impact, with resolution controls and a full history.

Optional AI explanations describe a finding in plain English using only the verified values — and say when a cause requires review rather than guessing.

Click Review Evidence on any finding to open the source document at the exact page. When position data exists, the value is highlighted on the page; otherwise Clariose shows the extracted source text.

Numeric questions — “How much dividend income did I receive in 2025?” — are answered by deterministic database queries, then phrased in plain language. The numbers you see come from the query, not the model.

Document questions — “Who is the successor trustee?” — are answered from retrieved passages, and every citation is validated against the source page before it is shown. When the evidence is insufficient, Clariose says so.