Documentation

SmartKMS

An AI-powered knowledge management platform for universities — upload institutional documents and make them instantly searchable and queryable through natural language, entirely on-premise.

SmartKMS in 60 seconds: upload, AI search, Q&A, sensitive-content screening, colleges and roles, and analytics.
On-premise by design No data leaves the institution. AI models run locally via Ollama, files are stored locally via MinIO, and the database runs locally via PostgreSQL. No cloud subscriptions, no per-query API costs, no internet dependency after setup.

What SmartKMS Does

Staff, faculty, and administrators upload institutional documents — policies, guides, research papers, manuals — and the system makes them instantly searchable and queryable in plain language. Instead of digging through shared drives or email threads, a user types a question and gets a direct answer, cited from real documents.

Document Upload & Management

Drag-and-drop upload for PDF, Word, Excel, CSV and images. Every file is auto-tagged by the AI, and the viewer suggests which folder it belongs in based on what's already there. Uploads can be cancelled mid-processing at any time.

In-Browser Document Editing

Word, Excel, PowerPoint, and PDF files open and edit directly in the browser via ONLYOFFICE — no download/re-upload round trip.

AI-Powered Semantic Search

Queries are converted to vectors and matched by meaning, not just keywords — "how do students appeal a grade" finds the right document even without matching words.

Q&A Chat Assistant

Ask questions in plain language and get a direct answer with citations back to the source documents (Retrieval-Augmented Generation).

Multi-College Architecture

Each college's documents are isolated by default; admins can make documents public within a college, global, or shared with specific people.

Sensitive-Content Screening

Every upload is screened for leaked personal information and explicit imagery before approval, without flagging legitimate educational or historical content.

Reapproval Workflow

Uploaders of a rejected document can request reconsideration; admins and staff review requests from a dedicated queue.

Analytics Dashboard

See what topics are searched most, which queries return no good results, and who contributes the most knowledge.

Where to Start

Technology Stack

LayerTechnologyPurpose
Backend APIFastAPI (Python 3.11+)REST API, async request handling
Package ManagementuvEnvironment + dependency management; uv.lock pins identical versions locally and in Docker
DatabasePostgreSQL 16 + pgvectorStores all data + AI vector embeddings
Database MigrationsAlembicAsync schema migrations
File StorageMinIO (Docker)Stores uploaded files
Document EditingONLYOFFICE Document Server (Docker)In-browser editing for Word/Excel/PowerPoint/PDF
AI — Chat & SummariesOllama · qwen2.5:3bGenerates answers and document summaries
AI — Search EmbeddingsOllama · nomic-embed-textConverts text to vectors for semantic search
AI — VisionOllama · llavaOCR and image-based content screening (nudity/gore)
AI — Nudity Pre-CheckNudeNet (separate Docker microservice)Deterministic bounding-box nudity classifier, runs alongside the vision-model screen; isolated in its own container (nudenet_service/) for AGPL licensing and called over HTTP
FrontendVanilla JS + HTML + CSSWeb UI served directly by FastAPI
AuthenticationJWT + httpOnly refresh cookieSecure stateless sessions
On this page