Financial baselines are built from the statements as the source documents print them, Excel files included, and export in a due-diligence report layout.
Source statements (決算書) view: statements and 内訳書 schedules exactly as printed, each figure traceable to its file, page and printed label.
An item reported in several documents is shown as one line, with each document it came from.
Excel sources: sheets are mapped to statements, periods are read by column, and rows hidden in the workbook are marked.
Valuation: the DCF says when a value is not meaningful instead of printing a figure, flags an enterprise value far from the market range, and shows the basis for each forecast driver.
DD-format workbook: a 正常収益力 sheet with normalized EBITDA by period, adjusted net assets (実態B/S) on one sheet, and a net-cash sheet with tax-effected bridge items.
The equity bridge proposes items such as 倒産防止共済 and 資産除去債務 for the analyst to accept.
Pages scanned sideways or upside down are turned upright before they are read.
Improvement
Cross-Page Tables, Two-Page Spreads & Tables to Excel
A table split across pages comes back as one table, and every table in a result can be downloaded as an Excel workbook.
A table that continues onto the next page is merged back into one table. Where the boundary is unclear, the model decides from both pages.
Period columns in born-digital financial statements are read from the PDF’s own text layer, so the figures are not retyped.
Scanned two-page spreads are read as two pages, in reading order.
Download a result’s tables as one .xlsx: a Contents sheet, then one sheet per table with its title and the text around it.
Pages are split into evenly sized batches, so a long document no longer waits on a small final batch.
Improvement
Every Excel Format Read from Its Cells & Office Previews
Older Excel formats are read from their cells instead of as page images, and Word and PowerPoint files are shown beside their result.
.xls, .xlt and .xlsb workbooks are read from their cells, keeping the grid, instead of being converted to page images and read as images.
Plain-text and Markdown files pass straight through without conversion.
Word, PowerPoint, OpenDocument and RTF uploads show the source document side by side with the result.
An office-document conversion that stalls is stopped at a time limit, so it no longer leaves a job stuck.
v1.13.0Feature
Answer-First Valuation, Transaction Perimeter & Version History
Valuation opens on the answer and runs on its own, the deal perimeter can be defined, and any earlier baseline can be restored.
The Multiple and DCF tabs open on the answer — enterprise value range, equity value and value per share — with the working one click away.
Valuation runs automatically once the baseline is generated, and estimates and evidence gaps are flagged beside the number instead of hiding it.
One status vocabulary across valuation — Specter estimate, Analyst input, Market source, Evidence gap — and WACC and terminal assumptions are edited in place.
The valuation workbook opens on a summary sheet, with the DCF forecast laid out by year and the equity bridge beneath it.
Transaction perimeter: define the business being acquired, the operations included and excluded, and how shared items are allocated. Changing it clears statements built on the old perimeter.
Version history: open any earlier baseline in full and restore it as the current one. A valuation can be regenerated in a chosen language, and a generation in progress can be cancelled.
Projects show whether they are yours, your organization’s or shared with you, with a filter to match; anyone a project was shared with can leave it, and the Data Room’s “Added by” column names who added each file.
Feature
Import Folders from Google Drive, OneDrive, Box & Dropbox
Link a cloud folder and Nebula converts its files as one batch.
Connect Google Drive, OneDrive, Box or Dropbox, pick a folder, and its files are converted as a batch.
“Import from cloud” sits beside Upload files and Upload folder.
Imported files are converted with the same settings as files uploaded by hand.
A file identical to one already converted is marked “Reused result” in the folder view and on the job page.
v1.12.0Feature
Organizations, Sharing & Baseline Review
Share projects across an organization, and accept baseline figures a reporting period at a time.
Organizations: access decisions now resolve in one place, and sharing a project works end to end.
Financial baseline: accept a whole reporting period at once instead of line by line.
The tracking list replaces the old export view, with per-row fulfillment status and pinned rows on by default.
Announcement
SOC 2 Type 2 and GDPR Compliance
Ur AI completed its SOC 2 Type 2 examination and GDPR compliance programme, covering the controls Specter runs on.
Audit logging, access control and admin accountability were built to the SOC 2 criteria rather than retrofitted to them.
GDPR data-subject rights are served by real endpoints — access, erasure, portability and records of processing.
The Trust Center at trust.ur-ai.net carries the current reports and control status.
Password-protected uploads wait for a password instead of failing, and pages are turned the right way up before they are read.
A password-protected upload parks as “needs password” and waits at an unlock step, instead of failing as unreadable.
Pages are checked for orientation before conversion and corrected, with a backup model when the primary one fails.
Excel template formats (.xltx, .xltm) go through the synchronous fast lane.
Spreadsheets carrying images are routed to the full pipeline, so nothing embedded in them is lost.
The upload allow-list accepts every file type Specter handles.
v1.11.0Feature
Cloud Data Room Connectors & the Equity Bridge
Import straight from Google Drive, Dropbox, Box or OneDrive, and bridge enterprise value to equity value with diagnostics you can audit.
Connect Google Drive, Dropbox, Box or OneDrive to a project and import into the Data Room with the cloud folder structure preserved.
Sync is on demand, not continuous: a freshness badge shows when a linked folder has changed, and you press Resync. A re-import skips any file whose content hash proves nothing changed, and every file that fails records why.
Unlink a cloud folder whenever you like — the files already imported stay where they are.
Equity bridge: a balance-sheet bridge from enterprise value to equity value, with its own table, diagnostics and export.
Valuation now covers negative-EBITDA cases through an EV/Sales route, a net-asset-value floor, and an explicit build-options view.
Staleness banners say when a projection or valuation no longer reflects the figures underneath it.
v1.10.0Feature
Report Export & Grounded Research
Export a finished report as Markdown, Word or PDF, and see exactly which research run produced what you are reading.
Export a report as Markdown, Word or PDF, with an executive summary generated only from what the report itself contains.
Research carries provenance at the run level: a report says which research produced it, and a banner appears when research it expected is missing.
Red Flags can generate its own structure when a project has none, gated on what that project actually supports.
A notification says “ready with gaps” when a step did not run, instead of reporting plain success.
One confirm dialog across the application; the browser’s native dialogs are no longer used.
Uploads are checked against the allowed content types in the browser, and server-side rejections are surfaced rather than swallowed.
v1.9.0Feature
Valuation Baseline, Comparables & Analysis Language
Driver-level projections that fade to a terminal rate, comparables you can regenerate deliberately, and one resolved language per analysis.
Financial baseline: driver-level projections that fade to a terminal rate, approved EBITDA adjustments shown in both the waterfall and the table, and diagnostics that explain a valuation’s sign.
Comparables: complete target profiles with an explicit regenerate action, published under a fencing token so two regenerations cannot collide.
One analysis language is resolved at every baseline entry point, and the prose the interface writes is localized with it.
Email notifications for structure and classification milestones, grouped into a section digest rather than sent one at a time.
A step that legitimately finds nothing to do is treated as done, so the steps depending on it are no longer left blocked.
Trade speed against accuracy per job, and read each region of a page with an instruction chosen for that region.
Choose reasoning effort per job — maximum speed at one end, maximum accuracy at the other — and the choice persists for your session.
Layout-aware crop reading: the page is segmented, each region cropped, and each crop read with an instruction resolved for that region.
Handwritten regions are routed to the model that reads them best.
Images take a fast path straight to the vision model, skipping the render stage they do not need.
A distributed rate governor keeps concurrency dense without tripping provider limits, with per-model rates.
Model coverage extended across current-generation GPT and Claude models, each with effort variants.
Announcement
Nebula is publicly available
Nebula opened to anyone with an account: turn Japanese and English business documents into layout-preserved Markdown and structured JSON, in the browser or over the API.
Charts, tables, forms, presentations and scanned pages convert with their structure intact.
Japan-hosted, with per-tenant isolation. Documents are not used to train models.
Available in the browser at nebula.ur-ai.net and over an authenticated API.
v1.8.0Feature
Notifications, Encrypted Files & a Faster Data Room
Know the moment work finishes, open password-protected files without leaving the app, and wait less for large libraries.
A notification bell with live arrival toasts and deep links into the work that just finished.
Password-protected files are detected on upload and opened with a password prompt, instead of failing as unreadable.
Batch uploads are fault-isolated: one file’s failure no longer takes the batch with it, and interrupted uploads retry.
Ingestion carries a completeness invariant, so financial rows are never silently dropped.
PowerPoint and Markdown files preview inline, and legacy .doc uploads work end to end.
A broad performance pass across ingestion and retrieval — parallel table fetches, pooled CPU work, and PDFs rendered in page ranges from a single open.
Feature
Accounts, API Keys and Prepaid Credits
A customer portal with real accounts, keys you control, and prepaid credits that stop rather than surprise you.
A customer portal with email accounts.
Prepaid credit billing through Stripe, with a hard stop when the balance runs out.
A pricing page with tiers, plus referral codes redeemable after signup.
Every charge is recorded in a usage ledger you can export.
Read-only billing endpoints available with an API key.
Feature
Custom Prompts and Per-Job Feedback
Tell us what a conversion got wrong on the job that got it wrong, with custom prompts managed in one place.
A per-job feedback thread in a floating widget — the report stays attached to the job it is about.
Custom prompts managed from a central store rather than edited per deployment.
Presentations convert through PDF, so slide decks keep their layout.
v1.7.0Feature
Conversational Analysis Rebuilt, Spanish, and Session Security
A rebuilt conversation surface with a proper citation viewer, Spanish end to end, and sessions hardened to enterprise expectations.
Conversational analysis rebuilt: a citation viewer with page navigation and full-screen reading, sources grouped by kind, and Excel, Word, CSV and text rendered inline instead of downloaded.
Scope a conversation to part of the data room with a file picker, or query the generated reports instead — Specter’s own outputs appear as files you can filter on.
Chat sessions persist, with suggestions on the empty state and a rolling summary that keeps long conversations coherent.
Spanish added end to end, alongside Japanese and English.
Library management with versioning, a slide view for business analysis sections, and charts enriched in the background and regenerated selectively.
Sessions moved to httpOnly cookies with CSRF protection, rate-limited authentication, and detection of refresh-token reuse.
GDPR data-subject endpoints (Articles 15, 17, 20 and 30) and SOC 2 logging, monitoring and admin accountability.
v1.6.0Feature
Shared Engine, Project Dashboard & Shared Links
Analysis moved onto the same document intelligence engine Nebula runs on, with a project dashboard and links you can send outside the account.
Analysis generation moved onto the shared engine Specter and Nebula both build on, so a document is parsed once and reused across every analysis.
A project dashboard summarizing library state, processing progress and analysis coverage in one place.
Shareable section links, so a finished section can be sent to someone without an account.
A reference library selector, to scope an analysis to a chosen subset of documents.
A chat scope selector: query the data room, the generated reports, or both.
Red Flags export questions and DDQ request data, plus a project context dialog captured after upload.
v1.5.1Improvement
Search Variants, Pagination & Red Flags
Smarter Japanese search with automatic name variants, pagination across data listings, and improved Red Flags accuracy.
Search now generates kana, romaji, and abbreviated name variants automatically for more comprehensive results across Japanese data sources.
Pagination and search history added to Registry and other data-heavy views.
Prefecture-based filtering and ranking for Nenkin and Kanpo results.
Table of Contents sidebar with auto-scroll for long CDD reports.
Red Flags analysis prompts improved to reduce false positives.
v1.5.0Feature
Web CDD, Registry & Data Room Rebuild
Full Web CDD workflow, a new Registry module for Japanese compliance records, and a reorganized Data Room.
Web CDD: run competitor due diligence with citations, table of contents, summary views, and export to data room.
Registry: search Nenkin (pension registry) and Kanpo (government gazette) records for target companies.
Classification and structure tree redesigned — creating a project now auto-generates its structure.
Red Flags and Teaser Analysis now support running multiple independent analyses.
Data room reorganized into Uploads, Outputs, and Registry tabs with folder tree view.
v1.4.1Improvement
Presentations, Web Fetch & AI Models
Redesigned presentation generation, new web fetch capability, and expanded AI model options.
Dedicated presentations page for managing and downloading generated PowerPoint decks.
Web Fetch: pull content from any URL into your analysis and save results to the data room.
Choose from Claude, Gemini, Qwen, or OpenAI models for analysis.
Improved progress indicators for long-running operations like document processing and report generation.
Analysis outputs and web fetch results automatically added to the combined document library.
v1.4.0Feature
UI Refresh, Security & File Formats
Refreshed interface, two-factor authentication, and support for 10+ document formats.
Visual refresh across all pages with cleaner layouts and consistent navigation.
Two-factor authentication (TOTP) with trusted device option. New profile pages for session management and login history.
Upload DOCX, PPTX, TXT, HTML, RTF, images (PNG, JPG), and EML emails alongside PDF and Excel.
Password reset flow added. IP address and device tracking on login for security.
v1.3.1Fix
Excel Processing & Stability
Fixed Excel, XLS, and CSV processing to handle empty cells and formulas correctly. General stability improvements.
Formulas stripped from Excel files before analysis to avoid misleading content.
Trailing empty rows and columns trimmed to reduce unnecessary processing.
Improved error handling and timeout management for large document sets.
v1.3.0Feature
Web CDD & Analysis Language
Company research you can pull straight into the data room, and a language choice at every step of the pipeline.
Web CDD: AI-powered company research with findings you can import into your data room.
Language selection at each step of the analysis pipeline for Japanese, English, or other output languages.
v1.2.2Feature
Gemini & Processing Optimizations
Gemini as an alternative model for analysis, and faster document processing.
Gemini model option added alongside OpenAI for analysis.
Document processing optimized with batched status updates and reduced token usage.
Interdependency analysis across sections for deeper structural insights.
v1.2.1Improvement
Page Citations & Stability
Citations now link to specific page numbers in source documents. Processing reliability improvements.
Analysis citations include page numbers so you can verify sources directly in the original document.
Improved processing progress percentage accuracy.
Page-level document classification for more granular content categorization.
v1.2.0Feature
Data Room, Presentations & Teaser Analysis
Centralized data room for all project files, PowerPoint generation from analysis results, and Teaser Analysis.
Data Room: organize uploads, outputs, and downloaded files in a folder hierarchy with search, rename, and bulk operations.
Generate and download PowerPoint presentations from your analysis — per section or full report.
Teaser Analysis for quick preliminary assessments, with timestamped history.
v1.1.1Improvement
Parallel Uploads & Progress Tracking
Upload multiple files simultaneously with real-time progress tracking. Audit log sorting and streaming fixes.
Files now upload in parallel for significantly faster library setup.
Progress bar shows document processing status in real time.
Audit logs sortable by date. Streaming connection stability improved.
v1.1.0Feature
Due Diligence & Business Analysis
Full analysis pipeline — upload documents, generate a report structure, classify content, analyze each section, and export formatted reports.
Business Analysis module with end-to-end workflow: structure generation, document classification, section-by-section analysis, and report view.
Editable structure tree — add, remove, and reorder sections before running analysis.
Analysis results appear as they are generated instead of waiting for the full run to complete.
Export results as Excel spreadsheets or download as ZIP.
v1.0.1Improvement
Downloads, Dedup Audit & Early Polish
Download results as Excel or ZIP, deduplication audit logging, and initial UX improvements.
Download analysis results as Excel spreadsheets or ZIP archives.
Deduplication audit log tracks every duplicate detection decision.
Privacy notice added. File upload simplified with better error messages.
Same-name file and folder upload handling. .DS_Store files automatically rejected.
v1.0.0Announcement
Platform Launch
Initial release with project management, document upload with deduplication, and user authentication.
Create and manage projects with file upload supporting PDF, Excel, and CSV.
Automatic duplicate detection before upload using file hashing.
User authentication with email verification, login, and protected routes.