Releases: xt0n1-t3ch/Reva
Reva v1.4.0 — real-time processing, guided showcase, web-app focus
Reva is a web application for reinsurance document intelligence: ingest, extract, cite, reconcile, review, and export.
Highlights
- Real-time processing stream — a live view that shows each source line as it is read, then each extracted field, schema mapping, and reconciliation, on a stage-by-stage timeline. Documents without OCR stream their parsed source text.
- Guided Showcase — a header-launched tour of every capability, with a one-click action to load the bundled demonstration dataset, so a presentation needs no external files.
- Review template view — a Source / Template toggle that renders each record in a clean, standards-ordered layout.
- Assistant date/time tool and reasoning forwarded when the provider emits it.
- Single-file Reva.exe that serves the packaged UI and API and opens in the browser, branded with the Reva application icon.
Stack
Next.js 16 + React 19 + Tailwind v4 frontend, ASP.NET Core .NET 10 minimal API, EF Core on SQLite (SQL Server configurable), multi-provider model support (local Ollama, OpenAI-compatible, HuggingFace cloud).
Changed
- Refocused on the web app; the retired desktop shell was removed.
- Professional-language pass across the interface and documentation.
Fixed
- The assistant no longer renders an empty message bubble.
- The work queue requests a page image only for formats that render one.
Full notes: see CHANGELOG.md.
Reva v1.3.0
Reva v1.3.0 — first stable AI-native release
Reva now ships as a single self-contained Windows application. Double-click Reva.exe and the full product runs on http://localhost:5187: the web UI, REST API, offline OCR, reconciliation engine, and a grounded assistant — all from one localhost origin. No Node.js, no separate web server, no cloud account, and no API key.
Highlights
- All-in-one executable — one self-contained
Reva.exeserves the static Next.js cockpit, REST API, OCR engine, and assistant chat from a single localhost port. - Grounded assistant chat — a local, keyless model (Ollama
qwen3-vl) answers questions and runs tools over your real documents (list, inspect, reconcile, explain a field). It degrades to a clear message when no model is installed; everything else keeps working. - Source citations end to end — OCR captures per-line bounding boxes and the split-view review highlights the exact region each field was read from, scaling with zoom.
- Scanned-document OCR — images and scanned PDFs are rendered and extracted with real page coordinates via the bundled offline PaddleOCR engine.
- Export template editor — create, edit, duplicate, and delete templates with format-aware CSV/Excel/JSON download and live preview, including a Lloyd's CRS template.
- Settings depth — configurable reconciliation tolerance, an optional LLM-assisted extraction toggle (off by default), export defaults, and data management.
- Rebuilt interface — a dense Next.js operations workspace with a work queue, drag-and-drop upload, learned per-sender schema mapping, and a first-run guided tour.
Install
- Download
Reva-v1.3.0-win-x64.zipbelow and extract it. - Double-click
Reva.exe(orStart-Reva.cmd). - Open
http://localhost:5187and upload a reinsurance document.
Optional assistant chat:
winget install Ollama.Ollama
ollama pull qwen3-vl:8bNotes
- Extraction and reconciliation are deterministic and run fully offline by default; the local model never overrides validated figures.
- Inbound email support is file-based
.eml/.msg; live mailbox sync is not included in this release. - The packaged application binds to localhost only and ships no secrets.
See the changelog for the full list of changes.
Reva v0.1.2
Added
- One-click Windows launch: double-click Reva.exe to start the local server and open the browser.
Fixed
- Unsupported/non-reinsurance uploads are quarantined instead of shown as successful extraction.
- Review/export controls are disabled for unsupported documents.
- Dashboard and review UI were redesigned into a denser Reva operations cockpit.
Validation
- CI: https://github.com/xt0n1-t3ch/Reva/actions/runs/27412443500
- Local: dotnet format, build, tests, Python worker tests, package smoke, browser proof.
Reva v0.1.1
Reva v0.1.1
This release fixes the local cockpit regression, completes the Reva rebrand inside the app, and adds a plug-and-play Windows package.
Fixed
- Fixed
/api/documents/returning HTTP 500 on SQLite becauseDateTimeOffsetordering was being translated into SQL. - Replaced the old
ReActive Intelligencerail branding withReva Document AI. - Improved tablet navigation so rail links no longer stretch awkwardly across the top bar.
- Ensured packaged SQLite runtime data directories are created automatically.
Added
Reva.exeWindows self-contained package inReva-v0.1.1-win-x64.zip.Start-Reva.cmdlauncher that openshttp://localhost:5187./healthendpoint for local/package verification.- In-process parser fallback for TXT, Markdown, CSV, PDF, and image visible-text intake when Python is not installed.
- Real package smoke test that builds the ZIP, extracts it, starts
Reva.exe, verifies/health, verifies/api/documents/, and shuts it down.
Validation
- GitHub CI passed on
fca99f7. - Local full gate passed: format, build, unit tests, integration tests, Python worker tests, package smoke, and diff check.
- Local app verified at
http://localhost:5187/#intakewith/health200 and/api/documents/returning extracted sample data.
Reva v0.1.0
Reva v0.1.0
Initial open-source release of Reva, a local-first AI document intelligence cockpit for reinsurance workflows.
Highlights
- Blazor analyst cockpit for upload, extraction review, exception triage, and export.
- ASP.NET Core API with SQLite-backed document workflow state.
- Local Python parser worker with Docling adapter path and deterministic fallback parsing.
- Reinsurance field extraction for technical accounts, bordereaux, and statements of account.
- Contract schemas, sample documents, architecture docs, AI pipeline notes, and CI validation.
Quick start
dotnet restore Reva.slnx
dotnet test Reva.slnx
dotnet run --project src/Reva.Web/Reva.Web.csprojValidation
- GitHub CI passed on the release-prep commit.
- Local app health verified at
http://localhost:5187.