Safiullah-Rahu/knowledge-intelligence
Knowledge Intelligence: a multi-tenant, permission-aware enterprise RAG platform. Grounded answers with citations over documents, databases and SaaS sources, plus Text-to-SQL, agents and a live connections map. FastAPI, PostgreSQL/pgvector, Next.js.
What it solves
Organizations often have knowledge scattered across documents, spreadsheets, databases, and code. This makes it hard for team members to find answers quickly while respecting who is allowed to see what. The platform lets teams ask questions and get answers that are grounded in sources they are allowed to see, with clickable citations. It solves the problem of secure, reliable knowledge retrieval across many data sources.
How it works
The platform connects to various sources (documents, spreadsheets, databases, web pages, and code). Content is ingested, split into chunks, and indexed using vector search (pgvector) and keyword search (BM25). When a user asks a question, the system retrieves relevant passages from the authorized sources (respecting permissions), then generates an answer with citations. It also supports text-to-SQL, so users can query databases directly in plain language, and it includes agents that can request approval for certain actions. The platform is multi-tenant, meaning different organizations are isolated from each other.
Who it’s for
This is for organizations that need a secure, permission-aware knowledge base where team members can ask questions and get answers with citations across a wide range of data sources. It is designed for enterprises that handle sensitive data and need fine-grained access control.
Highlights
- Multi-tenant by design, using PostgreSQL Row-Level Security to isolate tenants.
- Fine-grained access control: 55 permissions, 11 predefined roles, custom roles, workspace scopes, deny-wins, and expiry.
- Serious RAG: supports 30+ file formats, 9 chunking strategies, hybrid dense/sparse retrieval, reranking, MMR, query expansion, prompt-injection screening, and citation verification.
- Text-to-SQL with schema discovery, validation, and approval workflows.
- CSV-AI heritage: every uploaded spreadsheet becomes a queryable dataset.
- Connections command center: a live map of data sources, indexes, models, and agents built from real events.
- Operations: job queue with retries, webhooks (HMAC-signed), scoped API keys, audit log, analytics, and Docker Compose with TLS.
Related
- Project
- Project
- Project
- Project