📖 v1.0 · July 2026

UniDocVerse Help Guide

Your private, local-first AI document intelligence platform — from installation to advanced workflows, everything you need to get started and go deeper.

🖥 macOS Desktop · DMG installer
🐳 Docker · Team / server mode
🔒 100% private · all AI is local

1What is UniDocVerse?

UniDocVerse is a private document intelligence platform that lets you upload, analyze, and ask questions about any document — entirely on your own machine. No document ever leaves your device. All AI processing runs on phi3:mini via Ollama, a compact local language model.

🔒
100% Private
All AI processing is local. Your documents never reach a third-party cloud.
🧠
AI-Powered Q&A
Ask natural-language questions. Get grounded, cited answers from your documents.
🔍
Semantic Search
pgvector finds documents by meaning — not just keywords.
🌐
30 Languages
Full UI and AI response support for 30 languages.
🏢
Multi-Vertical
Insurance, Legal, Healthcare, Finance, HR, Real Estate, Supply Chain — all in one platform.
🔗
MCP Integration
Connect Claude Desktop, Cursor, Zed, Windsurf, and Continue as AI clients.

Who It's For

RoleHow they use UniDocVerse
Insurance professionalsAnalyze policies, COIs, renewals, loss runs, commission statements
Legal teamsReview contracts, NDAs, lease agreements, regulatory filings
Healthcare adminsProcess medical records, lab reports, intake forms
Finance teamsParse bank statements, invoices, tax returns, expense reports
HR professionalsScreen resumes, manage handbooks, review performance docs
Real estateHandle appraisals, title deeds, disclosures
Supply chainTrack purchase orders, bills of lading, shipping manifests
AdministratorsConfigure the platform, manage users, enable industry modules

2Installation

Option A — macOS Desktop (DMG)

The DMG is a fully self-contained installer. It bundles PostgreSQL, Ollama, phi3:mini, and the FastAPI backend — no internet connection required after installation.

  1. Download unidocverse_0.1.0_universal.dmg from the downloads page.
  2. Open the DMG and drag UniDocVerse.app to your Applications folder.
  3. Double-click UniDocVerse in Applications to launch.
  4. On first launch macOS may show a security prompt — click Open.
  5. The app starts PostgreSQL, Ollama, and the backend in the background (≈15 seconds).
  6. The web interface loads in the app window. You're ready.
⚠️

Gatekeeper warning: If macOS shows "cannot be opened because the developer cannot be verified", go to System Settings → Privacy & Security and click Open Anyway. The app is signed with a Developer ID certificate.

What happens on first launch

1
PostgreSQL initializes
Portable Postgres starts on port 54321 — no system installation needed.
2
Database migrations run
All schema migrations (V1–V39) apply automatically via the built-in migration runner.
3
Ollama starts with phi3:mini
The bundled Ollama binary loads the pre-baked phi3:mini model — no download needed.
4
Backend starts
FastAPI backend launches on port 8000 and serves the Angular frontend.
5
App opens
The UI loads. On subsequent launches, startup takes only a few seconds.

Option B — Docker (Server / Team Mode)

Run UniDocVerse as a multi-container stack for shared team access or server deployment.

# Pull and start all services
docker compose up -d

# Or pull the pre-built all-in-one image
docker pull bommavj/unidocverse:latest
docker run -p 80:80 bommavj/unidocverse:latest

Open http://localhost in your browser. Services included:

ServicePortRole
PostgreSQL + pgvector54321Primary database with vector search
FlywayRuns schema migrations V1–V39 on startup
Ollama11434LLM server hosting phi3:mini
FastAPI backend8000API server and document processing
Nginx (Angular)80Serves the web UI and proxies API calls
ℹ️

On first Docker run, ollama-init pulls phi3:mini from the Ollama registry (≈2.3 GB). Subsequent starts use the cached model from the ollama_storage volume.

3First-Time Setup

Logging In

🔐

Default credentials: username admin · password admin — change these immediately after first login.

  1. Open UniDocVerse (desktop app or http://localhost).
  2. Enter your username and password, then click Sign In.
  3. You're directed to the Dashboard.

Setup Checklist

#WhereWhat to do
1Settings → PasswordChange the default admin password
2Settings → FeaturesEnable the industry modules your team needs
3Settings → UsersCreate accounts for additional team members
4Sidebar → Language selectorSet your preferred UI and AI language
5UploadUpload your first document to verify the pipeline works

Dashboard Overview

After login, the Dashboard shows your at-a-glance summary: total documents, active alerts, storage used, documents uploaded this week, a needs-attention list, a recent activity feed, and a document type distribution chart.

4Document Upload

Supported File Types

📄 PDF 📝 DOCX / DOC 📊 XLSX / XLS / CSV 🖼 PNG / JPG / TIFF 📧 EML / MSG 📋 TXT / RTF 🗃 JSON / XML 📽 PPTX / PPT

Maximum file size: 50 MB per file. Multiple files can be uploaded simultaneously.

How to Upload

  1. Click Upload in the left sidebar.
  2. Drag files onto the drop zone, or click Browse to select them.
  3. Watch each file's progress indicator — stages: Uploading → Extracting text → Running phi3:mini → Indexing.
  4. When status shows ✓ Analyzed, the document is ready.

What the AI Does Automatically

StageWhat happens
Text extractionOCR for scanned PDFs/images; native extraction for digital documents
ClassificationIdentifies document type (invoice, insurance policy, medical record, etc.)
SummaryGenerates a 3–5 sentence executive summary
Key pointsExtracts the most important facts, dates, figures, and obligations
EmbeddingsCreates a vector representation for semantic search
ChunkingSplits text into overlapping segments for precise Q&A retrieval
AlertsFlags expiring dates, anomalous amounts, missing required fields

Document Sheet — Five Tabs

Click any document to open its detail panel:

Summary

AI-generated executive summary with confidence score. Click Regenerate if unsatisfactory.

Key Points

Bulleted list of the most important facts extracted from the document. Click Load More for additional points.

Analytics

Structured data extracted for this document type — amounts, dates, parties, coverage limits, diagnoses, etc.

Ask AI

Full conversational interface scoped to this document. Answers stream word-by-word with confidence scores. All conversations are stored in History.

Details

Metadata (filename, upload date, type, size), document preview, download, delete, and related documents via semantic similarity.

5Ask AI

Ask natural-language questions about any document — or across your entire library simultaneously.

Per-Document Questions

Open a document → click the Ask AI tab. Questions are answered using only that document's content. The AI adapts its role to the document type:

Document typeAI role
Insurance policyCoverage expert
InvoiceAccounts payable specialist
Contract / NDALegal document assistant
Medical recordMedical records specialist
Bank statementFinancial analyst
Any other typeGeneral document assistant

Global Ask AI

Click Ask AI in the sidebar navigation to search across all documents simultaneously. The system retrieves the most relevant passages from any document and generates a unified answer.

How Answers Are Generated

  1. Your question is converted to a vector embedding.
  2. pgvector finds the most semantically relevant text passages.
  3. phi3:mini generates an answer grounded only in those passages.
  4. The answer streams token-by-token as it is generated.
  5. A confidence score is calculated and displayed.

Hallucination-resistant: The AI only answers from content in your documents. If an answer cannot be found, it will say so rather than fabricate information.

6Semantic Search

The Search page uses pgvector to find documents by meaning, not just keywords. Searching "when does my coverage end?" finds policies with expiration dates even if those exact words don't appear in the document.

How to Search

  1. Click Search in the sidebar.
  2. Type a natural-language query in the search box.
  3. Results appear as you type (400ms debounce).
  4. Each result shows a match score≥70% Excellent 50–69% Good <50% Fair.
  5. Click Open → on any result to view the full document.
💡

Use the Suggested Queries chips on the Search page for common industry-specific queries pre-optimized for your active modules.

7Conversation History

Every question you ask about any document is stored and searchable in the History page.

Layout

Left panel — all documents you've had conversations with, showing the most recent question, turn count, and time elapsed. Right panel — the full Q&A thread for the selected document, with questions in indigo bubbles and answers in white bubbles with confidence scores.

Searching Past Conversations

  1. Click History in the sidebar.
  2. Type in the search bar at the top of the sidebar panel.
  3. Matching Q&A pairs appear — click one to jump directly to that turn in the thread, highlighted.

Deleting Conversations

Hover over any Q&A card in the thread to reveal the delete button. Confirm in the prompt to remove that exchange. Use Load More at the bottom of the thread to paginate older conversations.

8Compliance Hub

Click Compliance in the sidebar to access three tabs: Deadlines, Alerts, and Audit Trail.

Alerts

Proactive alerts generated by document analysis, sorted by severity:

🔴 Expired 🟠 Critical — <30 days 🟡 Urgent — <60 days 🟣 Warning — <90 days

Filter by severity tab, search by filename, and click Open → to navigate directly to the source document.

Deadlines

All upcoming compliance-related deadlines extracted from documents, sorted by urgency. Mark deadlines as complete when resolved.

Audit Trail

Complete log of all system actions — uploads, deletions, questions asked, setting changes — with timestamp, user, and action type. Fully searchable.

9Analytics

The Analytics Hub provides collection-level insights about your document library: total document counts, type distribution, processing statistics, storage usage trends, and alert volume over time.

For per-document analytics, open any document and click the Analytics tab in the document sheet to see structured data extracted specifically for that document type.

10AI Agent

The AI Agent uses a ReAct (Reason + Act) loop to perform multi-step tasks across your document library. It can search documents, retrieve client data, analyze policies, and compile reports — autonomously, with your approval for sensitive actions.

Starting an Agent Run

  1. Click Automation in the sidebar, then the Agent tab.
  2. Type a task description (e.g., "Find all policies expiring in the next 30 days and summarize the coverage gaps").
  3. Click Run. The agent begins its reasoning loop.
  4. Each step is shown in real-time in the Agent Queue panel — tool calls, reasoning, and results.
  5. If the agent needs your approval for an action, it pauses and shows an approval card.
🤖

The agent uses phi3:mini for reasoning and has access to all UniDocVerse tools: document search, client lookup, policy retrieval, analytics, and more.

11Document Research Agent

The Document Research Agent is a multi-step AI reasoning engine that answers complex questions by autonomously searching, reading, and synthesizing information across your entire document library. Unlike single-document Q&A, it can cross-reference multiple documents, filter by type, and produce a comprehensive cited answer in one shot.

🧠

Find it under Analytics & Insights → Document Research Agent in the sidebar.

How It Works

The agent runs a ReAct loop (Reason + Act) — it thinks through your question, decides which tool to use next, observes the result, and repeats until it has enough evidence to write a final answer. Each step is shown in real time.

StepToolWhat it does
🧠Understand QuestionBreaks down your question into a research plan before taking any action
🔍Search DocumentsRuns a semantic vector search across your library to find the most relevant passages
💬Ask DocumentAsks a targeted question to a specific document and retrieves a grounded answer
📋Filter DocumentsLists and filters documents by type, date, or keyword to narrow the search space
Final AnswerSynthesizes all gathered evidence into a single, cited, comprehensive response

Running a Research Query

  1. Navigate to Analytics & Insights → Document Research Agent.
  2. Type your question in the research input — or click one of the example chips to get started.
  3. Click Research. The agent begins its reasoning loop immediately.
  4. Watch each step appear in real time: tool name, its reasoning, and the result it observed.
  5. Click any step card to expand it and read the full detail of that reasoning step.
  6. The Final Answer appears at the bottom once all evidence is gathered.

Example Questions

💡 What are the main risk factors for this company? 💡 How does executive compensation compare to performance? 💡 What is the pipeline strategy and timeline? 💡 Why is the company proposing to increase authorized shares? 💡 What are the key revenue drivers and financial trends?

Research History

Every research run is saved automatically. Access past research sessions from Analytics & Insights → Agent Conversation History. The history panel shows all prior questions in the left sidebar — click any entry to reload its full step-by-step reasoning and final answer.

Best for: complex multi-document questions, competitive analysis, portfolio reviews, compliance checks across many files, or any question that requires reading more than one document to answer fully.

12Industry Modules

UniDocVerse ships with optional industry modules enabled by an administrator in Settings → Features. Each module unlocks a set of sidebar sections and document hub tabs.

ModuleUnlocks in sidebar
🏢 Agency ModeClients, Policies, Renewals, Commission, Loss Runs, COI, Quotes, Claims
⚖️ Legal ModeMatters, Contracts (CLM), NDAs, Lease Agreements, Regulatory Filings, Approvals
🏥 Healthcare ModeMedical Charts, Intake Forms, Lab Reports, Risk & Health
💰 Finance ModeBank Statements, Tax Returns, Invoices
🏠 Real Estate ModeAppraisals, Title Deeds, Disclosures
👥 HR & People ModeResumes, Handbooks, Performance Reviews
📦 Supply Chain ModePO Register, Bills of Lading, Manifests
✅ Compliance & RiskCompliance, Risk, Audit, and Deadlines pages
⚙️ Operations ModeMatters, Claims, Approvals, Quotes, CLM, Jurisdiction

Enabling a Module

  1. Log in as an administrator.
  2. Go to Settings → Features.
  3. Toggle the desired module On.
  4. The sidebar immediately shows the new section — no restart needed.

13MCP / AI Integration

UniDocVerse exposes a Model Context Protocol (MCP) server at http://localhost:8000/mcp/sse. Connect it to AI coding tools to query your documents and run UniDocVerse tools from inside your AI assistant.

🔴

Important privacy notice — please read before enabling MCP.

UniDocVerse is designed to be 100% private and local. Your documents never leave your machine when you use the app normally. MCP integration changes this.

When you connect an AI client (such as Claude Desktop, Cursor, Zed, Windsurf, or Continue), that client can query your UniDocVerse documents on your behalf. Every query — including document text, extracted summaries, key points, client names, policy numbers, financial figures, medical details, and any other content in your library — is sent to that AI client. The AI client then forwards this data to its own cloud servers (for example, Anthropic's servers for Claude Desktop, or the respective vendor's servers for other clients).

What this means in plain terms:

  • Your document content may be used to train or improve third-party AI models, depending on that vendor's data policy.
  • Sensitive PII (names, addresses, SSNs, dates of birth) in your documents may be transmitted to and stored on external servers.
  • Protected health information (PHI), privileged legal communications, and confidential financial records are equally at risk.
  • Once data leaves your machine via the AI client, UniDocVerse has no control over how it is handled, retained, or shared.

Recommendation: Only enable MCP integration if your document library contains non-sensitive or already-public information, and you have reviewed and accepted the privacy policy of the AI client you are connecting. If you handle regulated data (HIPAA, GDPR, SOC 2, attorney-client privilege), do not enable MCP.

Auto-configure: Open UniDocVerse → Settings → Integrations → click Connect next to any supported client. UniDocVerse automatically writes the correct config file for you.

Supported Clients

Claude Desktop
~/Library/Application Support/Claude/claude_desktop_config.json
{ "mcpServers": { "unidocverse": { "url": "http://localhost:8000/mcp/sse", "type": "sse" } } }
Cursor
~/.cursor/mcp.json
{ "mcpServers": { "unidocverse": { "url": "http://localhost:8000/mcp/sse", "transport": "sse" } } }
Zed
~/.config/zed/settings.json
{ "context_servers": { "unidocverse": { "command": { "url": "http://localhost:8000/mcp/sse" } } } }
Windsurf
~/.codeium/windsurf/mcp_config.json
{ "mcpServers": { "unidocverse": { "serverUrl": "http://localhost:8000/mcp/sse", "transport": "sse" } } }
Continue
~/.continue/config.json
{ "experimental": { "modelContextProtocolServers": [{ "transport": { "type": "sse", "url": "http://localhost:8000/mcp/sse" } }] } }

What You Can Do via MCP

Once connected, your AI client can call any UniDocVerse tool — search documents, query clients and policies, retrieve analytics, manage approvals, run the AI agent, and more — directly from within your AI assistant's chat or code completion interface.

14Admin & Settings

The Settings page is only visible to administrators. It has four tabs:

Users Tab

ActionHow
Add a userFill in full name, username, password, and role (Admin / User) → click Add User
Disable a userToggle the active switch next to their name
Delete a userClick the trash icon (permanent — cannot be undone)
Reset passwordClick the user row and enter a new password

Features Tab

Enable or disable industry modules, compliance/risk, and operations sections. Changes take effect instantly with no restart required.

Password Tab

Change your own account password. Enter current password, new password, and confirm. Minimum 8 characters.

RBAC — Roles

RoleCapabilities
AdminFull access — all pages, Settings, user management, feature flags
UserUpload, query, search, history, compliance — cannot access Settings

15Report an Issue

Found a bug, have a feature request, or need to ask a question? Use the Report Issue form built directly into UniDocVerse — no external tools or email required.

📍

Find it in the sidebar under Report Issue (available to all users).

How to Submit an Issue

  1. Click Report Issue in the left sidebar.
  2. Enter a clear Title summarizing the problem or request.
  3. Choose an Issue Type:
TypeWhen to use
🐛 BugSomething is broken or not working as expected
Feature RequestYou'd like a new capability added to the platform
ImprovementAn existing feature that could work better
QuestionYou need clarification on how something works
  1. Set a PriorityCritical High Medium Low
  2. Write a detailed Description — include steps to reproduce, what you expected, and what actually happened.
  3. Click Submit. You'll see a confirmation once it's received.

Who Can See Your Report?

🔒

Only administrators can view submitted issues. Regular users can submit reports but cannot see the issue tracker or any reports submitted by others. Your report goes directly to your platform administrator — it is not sent to any external service or the UniDocVerse team.

Administrators can view, manage, and act on all submitted issues from the Settings → Issue Tracker page.


16Troubleshooting

macOS says "UniDocVerse cannot be opened because the developer cannot be verified"
Go to System Settings → Privacy & Security. Scroll down to the "Security" section and click Open Anyway next to the UniDocVerse entry. This is a one-time confirmation per app version.
The app takes a long time to load on first launch
First launch initializes the database (runs all migrations) and starts Ollama with phi3:mini. This typically takes 15–45 seconds on Apple Silicon and up to 90 seconds on Intel Macs. Subsequent launches are much faster (5–10 seconds).
"Unable to reach phi3:mini" error in Ask AI
Ollama may not be running or the model hasn't loaded yet. Wait 30 seconds after app launch for the model to initialize. If the error persists, quit and relaunch the app. On Docker: run docker compose logs ollama to check if the model was pulled successfully.
Document stuck in "Processing" state
Large PDFs (50 MB) with heavy OCR requirements can take 2–5 minutes. If a document has been processing for more than 10 minutes, check the backend log at ~/Library/Application Support/unidocverse/logs/backend.log for errors. Encrypted or password-protected PDFs cannot be processed.
Docker services won't start — port conflict
UniDocVerse uses ports 80, 8000, 11434, and 54321. If another service is using one of these ports, stop it first or edit docker-compose.yml to remap the conflicting port (e.g., change "80:80" to "8080:80" for the frontend).
Industry module sections don't appear after enabling in Settings
After toggling a feature flag, the sidebar updates immediately. If it doesn't, try logging out and back in to refresh the feature flag state. Ensure you're logged in as an administrator — the Settings page is not visible to regular users.
MCP connection not working in Claude Desktop / Cursor
Make sure UniDocVerse is running (the backend must be on port 8000) before your AI client tries to connect. Use the auto-configure feature: Settings → Integrations → Connect. Then restart your AI client to pick up the new config. Verify the config was written by checking the file path for your client listed in the MCP section above.
Responses are slow — AI takes 30+ seconds to answer
phi3:mini requires ≈2.3 GB RAM during inference. 16 GB RAM is the minimum recommended — on machines with less, response times will be significantly slower and the system may become unresponsive under load. 32 GB is recommended for consistently fast performance, especially when running other applications simultaneously. Apple Silicon (M1/M2/M3) is significantly faster than Intel for inference.

17System Requirements

ComponentMinimumRecommended
Operating systemmacOS 13 VenturamacOS 14+ (Apple Silicon)
RAM16 GB32 GB
Storage10 GB free20 GB free
CPU8-core Intel Core i7 / i9 (10th gen+)Apple M2 / M3 / M4
Docker (server mode)Docker Engine 24+Docker Desktop (latest)
💡

The DMG is a universal binary — a single installer works on both Apple Silicon (arm64) and Intel (x86_64) Macs. No separate downloads needed.