What’s new

Changelog

Every user-visible change to DocuSift, newest first.

All notable changes to DocuSift are recorded here. The format follows Keep a Changelog and the project adheres to Semantic Versioning.

1.0.0 — 2026-05-15

First public release of DocuSift — a multi-tenant document processing platform that classifies and extracts structured data from PDFs and images without per-vendor templates.

Document extraction

  • Template-free extraction using multimodal LLMs (Anthropic, OpenAI, Google) — classifies the document type and returns structured JSON in a single pass.
  • Accepts PDF, JPG, PNG, and TIFF up to 30 MB; scanned, native, handwritten, and multi-page documents all run through the same pipeline.
  • Two confidence scores per document (classification + extraction) with a per-tenant auto-approve threshold — above the bar auto-approves, below it lands in the Review queue.
  • Review queue with side-by-side PDF + editable fields, duplicate detection via content hash, and a processing-attempt timeline with exponential-backoff retries.
  • Ingest sources: drag-and-drop upload, mobile camera capture, email intake with SPF/DKIM/DMARC enforcement, and the public REST API.

Authentication & accounts

  • Email + password signup with mandatory verification; password reset via time-limited tokens.
  • Multi-factor authentication (TOTP) with recovery codes, lockout after repeated failures, and step-up challenges on sensitive actions.
  • OIDC SSO with auto-provisioning and role mapping (per-tenant SsoProvider).
  • Social login via Google and Microsoft built on the same OIDC primitives.
  • Session management with JWT refresh tokens, device tracking, and per-session revocation.
  • GDPR: one-click personal-data export (JSON) and self-service account deletion, both audited end-to-end.
  • Email-verification enforcement on sensitive actions (invites, API keys, billing mutations, integration OAuth) with a dismissible nag banner carrying a rate-limited resend.

Billing & plans

  • Multi-gateway checkout across Stripe, Razorpay, and Dodo Payments so tenants can use the gateway best suited to their region.
  • Free, Pro, and Enterprise plans with per-plan feature flags and page quotas.
  • Mid-cycle plan upgrades and downgrades with prorated usage accounting.
  • Page-level usage metering feeding the admin analytics dashboard.
  • Webhook-driven payment-event ingestion with idempotent processing.

Integrations

  • QuickBooks Online OAuth + invoice/bill push with field mapping and sandbox validation.
  • Xero OAuth + data push through the same routing-rule engine.
  • AWS S3 storage with presigned URLs; tenants can optionally bring their own bucket with per-tenant credentials.
  • Outbound webhooks with HMAC-SHA256 signatures, exponential-backoff retries, and a per-tenant sync-history drawer.
  • Routing rules for conditional auto-sync (e.g. only push invoices over a threshold to QuickBooks).
  • Trash / restore workflow for documents and integrations so destructive actions stay reversible.

Public API

  • POST /api/v1/extract for multipart document upload and GET /api/v1/extract/:id for polling status + extracted payload; every response follows the { data: ... } envelope.
  • API-key auth with per-key rate limits and a rotation workflow.
  • Webhook push on document.processed with signed payloads for event-driven integrations.
  • OpenAPI schema + Swagger UI at /docs for integrators.

Admin & tenant management

  • Multi-tenant isolation with super_admin, tenant_admin, and tenant_user roles.
  • Admin growth dashboard: signup funnel, active tenants, plan distribution, page-count trend.
  • Per-tenant audit log queryable by actor, table, and action.
  • Tenant suspension with structured reason codes and an internal TenantNote stream for ops context.
  • Invite / role / remove flows for team members with audit coverage.

Marketing & public site

  • Landing page with animated hero, pricing, testimonials, and a developer section with copy-ready upload / poll / webhook code samples.
  • Lead magnets — ROI calculator, free extraction audit, sandbox request, and template-free extraction guide PDF — with UTM attribution carried through to signup.
  • Public pages: pricing, contact, terms, privacy, and this changelog, each with per-page SEO metadata and an auto-generated sitemap.
  • Interim beta gate on public signup; scheduled to be removed on 2026-05-15.

UX & design system

  • Light and dark themes driven entirely by CSS custom properties in frontend/src/index.css; [data-theme='dark'] switches the whole surface.
  • Token-first CSS — every component consumes var(--*); token-check and style-check scripts enforce the rule in CI.
  • i18n scaffolding (react-i18next) with English as the default locale and a drop-in path for additional locales.
  • Reusable primitives: toast notifications, promise-based confirm dialogs, drawers, onboarding wizard, PDF viewer, markdown renderer.
  • Mobile-first responsive layout tested at 320 / 768 / 1024 / 1440 viewports; WCAG 2.1 AA contrast and keyboard reachability.

Observability & compliance

  • Structured Pino logs with request-ID correlation across the stack.
  • Per-route rate limiting via @fastify/rate-limit with stricter limits on auth and API-key routes.
  • Security headers via @fastify/helmet; CORS configured per origin.
  • LLM call logging with token counts and cost attribution (LlmCallLog).
  • Per-tenant data-retention policy (default 365 days) with a deletion scheduler.
  • Consent tracking and privacy-by-design data model — secrets (password hashes, MFA keys, token hashes, OAuth tokens) never leave the server.

Developer experience

  • TypeScript end-to-end across the Fastify backend and the React 19 / Vite 7 frontend.
  • Prisma 6 with auto-generated migrations and a seed workflow for local bootstrapping.
  • Vitest + Testing Library with axe-core accessibility assertions wired into component tests.
  • ESLint + Prettier + token-check + style-check quality gates bound to the standard task loop.
  • pg-boss background-job queue for async work (extraction, sync, email retries).