Features
Everything that ships in the private beta
Not a chatbot and not a search box. A tool for locating evidence across unstructured documents, with every citation bound to the claim that uses it and a person deciding what counts.
Evidence Programmes
Requirements in, reviewed findings out
- Evidence Programmes
- Run a named subject (a supplier, a site, a contract) against a set of requirements and get one finding per requirement, with its evidence attached.
- Supplier Compliance Review template
- A starting set of reviewer-written requirements, in the language a compliance reviewer already uses.
- Your own requirements
- Write requirements in your own words. They persist between runs, so the same question is asked the same way next quarter.
- Evidence states
- Each finding keeps passages DataStruct could support apart from sources it relied on but could not confirm, and from material that is merely related.
- Reviewer decisions
- Confirm, reject or mark for follow-up. Attributed to a named person and audit logged. The system records no verdict of its own.
- Runs and history
- Every run has its own identity, date and status. Open an earlier run to see what was found and decided then; decisions never carry over to a new run.
- Requirement editing and archiving
- Correct wording for future runs while earlier findings keep the wording they were judged against, or archive a requirement so future runs skip it.
- Programme reports
- Export one run as DOCX, XLSX, CSV or JSON: requirements, findings, evidence states, reviewer decisions and sources, with the human-review boundary stated.
Finding evidence
Found when the words do not match
- Hybrid search
- Keyword, semantic and section signals merged, so evidence is found when the document does not use your vocabulary.
- Page and section locators
- Citations name the document, the page and the section heading. A heading the cited text contradicts is withheld rather than shown.
- Spreadsheet evidence
- XLSX registers and trackers are first-class documents. Evidence names the sheet and the row or row range it came from.
- OCR for scanned documents
- Scanned PDFs are read rather than silently ingested as empty, and the deployment reports whether OCR is available.
- Search and ask
- Ask a question across your documents. Claims link to the passages they cite; a verification pass flags claims it cannot match to the evidence.
- Bulk upload and deduplication
- Upload up to 50 files at once. Duplicates are recognised by content hash, not file name.
Extraction and rules
Structured data with its source attached
- Extraction templates
- Define fields with types and validation, then run them across any number of documents. Each value carries a confidence score and a source link.
- Rule packs
- Deterministic checks: keyword presence and absence, regular expressions, numeric thresholds and missing-disclosure checks, each with a configurable severity.
- Document comparison
- Compare documents side by side and build peer groups to benchmark disclosures across entities.
- Document reports and datasets
- Document reports in HTML, DOCX or JSON; extraction datasets in CSV, XLSX or JSON, with source links preserved.
Access and audit
Controls an evaluator can check
- Workspace roles
- Owner, Admin, Member and Viewer, with granular permissions. Every workspace endpoint is permission-gated.
- Audit log
- Sensitive actions are recorded with the acting user and, where a value changed, its previous and new state.
- API keys
- Keys are stored hashed, can carry an expiry, and are revoked immediately. A REST API of 100+ endpoints, authenticated by session or API key.
- GDPR export and erasure
- A self-serve data export and account deletion in settings, with operator-completed deletion where self-serve cannot finish it.
- AI usage caps
- Per-organisation daily and monthly caps stop AI processing at the limit, and each AI call is logged.
- Industry configurations
- Terminology, document types, extraction templates and rule packs for 11 sectors, configured with you at onboarding.
Not available yet
Labelled plainly, so nobody has to ask
- Single sign-onNot available
- OAuth and OIDC exist in the codebase but are not proven in production, so beta accounts use email and password.
- Two-factor authenticationNot available
- Not available yet. Sign-in is email and password; Sign out everywhere ends every session and revokes API keys if a device is lost.
- WebhooksNot available
- Signed webhooks on document events are planned. The REST API is the integration surface today.
- PDF exportNot available
- Not implemented. Reports export as DOCX, XLSX, CSV, JSON or HTML.
- Customer-managed AINot available
- Bringing your own model credentials is on the roadmap. Today every AI call runs on DataStruct’s managed account.
Judge it on a real evidence set
Book a demo and bring the documents you would otherwise read by hand.