utopia: Build a Time-Aware Enterprise Knowledge System

Overview

Utopia is a self-hosted application for building an enterprise knowledge foundation from documents, data sources, and time-stamped observations. It combines document search with an ontology-guided graph that records both when a fact was true and when the system learned it, rather than simply replacing older information.

This makes Utopia suited to teams that need to answer questions with evidence, reconstruct what was known at a past date, or let agents read governed knowledge. It is a substantial system rather than a drop-in search library: deployments use PostgreSQL and require configured chat and embedding model endpoints for document extraction and chat.

Key Features

• Ingests common office documents and text, with scheduled syncs for sources including web pages, RSS, GitHub, Jira, Notion, WebDAV, and S3-compatible storage.
• Combines Tantivy full-text search and pgvector search, with answers that cite source passages.
• Stores facts in a bitemporal graph, preserving both their validity period and when the system learned them, along with their source evidence.
• Uses an editable ontology for extraction, entity resolution, conflict handling, and optional rule-based fact derivation.
• Routes uncertain extractions and conflicts to a human review queue, and records decisions in an append-only ledger.
• Provides a web interface and read-only MCP tools for agents, with knowledge-base membership and roles.
• Can connect to configured database sources so the agent can query structured data alongside documents.

Use Cases

• Legal and procurement teams can trace which contract terms applied on a particular date and identify the amendment and source passage behind them.
• HR, finance, and research teams can reconstruct who held a role or what the organization believed during a past period, including later corrections.
• Engineering and operations teams can give agents evidence-backed, date-specific access to a knowledge base through MCP, while keeping writes subject to review.
• Robotics, monitoring, or security systems can submit timestamped observations and query what an attribute was at a given moment.

Getting Started

With Docker installed, clone the repository and start the application:

git clone https://github.com/deeplethe/utopia.git
cd utopia
docker compose --profile app up -d

Open http://localhost:1516, register, then configure chat and embedding model endpoints under Administration ? Models before extracting documents. See the README for setup details and configuration guidance.

Alternatives

• KAG: KAG is a Python framework for domain-specific graph question answering, while Utopia is a self-hosted knowledge system with bitemporal records and human review.
• llm_wiki: LLM Wiki is a desktop app that maintains a linked wiki from documents, rather than Utopia's team-oriented, ontology-guided bitemporal graph.
• graphrag: TigerGraph GraphRAG combines graph and vector retrieval on TigerGraph; Utopia centers on an ontology-guided bitemporal graph and review workflows.
• graphrag: Microsoft GraphRAG is an indexing pipeline for graph-enhanced question answering, while Utopia provides a persistent knowledge system with agent access and review.

Considerations

• Utopia is at v0.1, and its database schema evolves between versions. Migrations only roll forward, so back up the database and data directory before upgrading and pin a version for production.
• The deployment requires Docker and PostgreSQL with pgvector. Document extraction and chat also depend on configured model endpoints; scanned documents and recordings may require additional vision or transcription services.
• The README advises reading SECURITY.md before exposing a deployment to the public internet.
• Ontology-driven derivation is off by default because incorrect axioms can produce incorrect derived facts. The project describes decision reasoning as in development.