Messy files in.
Validated data out.
Every run traced end-to-end.
APODEXA is a reliability layer that turns messy bank and text-based invoice files into validated, structured data for software and automation workflows — with a confidence score, validation checks, and full provenance on every run.
Works with the formats and tools you already use
Everything you need to trust a number
Deterministic parsing, honest confidence, and full provenance on every run — so your data is defensible, not just extracted.
Validation & confidence
Structural, type, and running-balance checks: each balance must follow from the row before it, so a missing, altered, or duplicated transaction is caught — not passed. Every field also carries a heuristic confidence score, surfaced honestly.
Learn moreFull provenance
Every run records where the data came from: parser, column mapping, and each transformation applied — plus a source-row reference for every normalized bank-statement row.
Learn moreReview workflow
Fields the parser isn't sure about route to a review queue instead of being silently guessed. You keep control.
Learn moreStructured recovery
Wrong encodings, mixed date formats, broken headers — recovered into a clean canonical ledger without dropping rows.
Learn moreEverything in one validated workspace.
APODEXA brings upload, validation, review, and export together — so you can focus on the numbers, not the cleanup.
- Real-time validation & confidence
- No-code review queue
- Provenance on every run
- Role-based workspaces
APODEXASame input. Same output. Every time.
The whole pipeline is rule-based code — no AI guesswork. Given the same file it always produces the same result, so you can reproduce, diff, and audit any output. Nothing is invented; anything uncertain is flagged, not fabricated.
Re-run the same file next month and get the same validated output, pinned by an output hash.
Uncertain fields go to review — never silently guessed or filled. A scanned or image-only PDF is refused outright, not guessed at.
Bank rows trace to their source row, and every run's record is ready to export for your audit trail.
Guarantees, by design: 0 AI guesses in the parsing path · 100% of runs carry full provenance · exact integer minor units — never floats · the same output hash on re-run (deterministic results) · 90-day retention
Every figure traces back to a source row, so month-end reconciliation doesn't have to be a manual re-check.
One endpoint, one response envelope — wire bank-CSV normalization into a pipeline and branch on structure, not surprises.
Uncertain fields go to a review queue instead of being guessed: clean data downstream, and an audit trail you can stand behind.
Straight answers about how it works.
Does APODEXA use AI or an LLM to read my files?
No. APODEXA parses files with deterministic, rule-based code — there is no LLM or machine-learning model in the processing path. The same file always produces the same output, so every result is reproducible and auditable. Uncertain values are flagged for review, never invented.
What file formats and banks does it support?
Bank and card statements as CSV — including Shift-JIS / CP932 Japanese exports with headers like 日付・摘要・出金・入金・残高 — and invoices as text or text-based PDF. It detects encoding, maps columns, normalizes ambiguous dates and locale-specific amounts, and outputs a canonical ledger with exact integer minor units, never floats. Scanned or image-only PDFs aren't OCR'd — they're detected and rejected, so you can supply a text export instead of a guess.
How is this different from an AI or LLM extractor?
LLM extractors are probabilistic: the same file can produce different answers on different runs, and the confidence they report is often fabricated. APODEXA is deterministic — identical input gives identical output, every normalized bank row traces to its source row, every run's provenance is recorded, and low-confidence fields go to a human review queue instead of being guessed.
What is provenance, and why does it matter for finance data?
Every run records how the result was produced: the parser and its version, the column mapping, and each transformation applied — plus, for normalized bank-statement rows, the exact source-row reference. That record exports as your audit trail, so a result can be defended in an audit or reconciliation instead of being a black-box guess.
Is my financial data sent to third-party AI services?
No. Files are processed on APODEXA's own infrastructure (AWS, Tokyo) with deterministic code. Nothing is sent to third-party AI or LLM providers, and we do not train models on your data. Data is retained for 90 days by default.
Is there a free plan and an API?
Yes. The Free plan processes 20 files per month at no cost and includes a sandbox API key (the same endpoints and shared quota, for evaluation). Paid plans (Starter $9/mo, Pro $39/mo) add higher volume, more templates, and full REST API access with a single consistent response envelope containing data, confidence, validation, and provenance.
Get a sandbox key and make your first API call → developer quickstart
Messy files in. Validated data out.
Process your first file free — every field scored and checked, every run traced.