Utopia is the world's first open-source enterprise world model: a self-hosted knowledge substrate with a bitemporal graph and an ontology at its base layer, built by DeepLethe for teams that need agent memory, governed knowledge, and a trustworthy decision core running on hardware they control.

What it is

Utopia is an open substrate for knowledge engineering that learns passively and governs itself. It is written in Rust, licensed under Apache-2.0, and lives in the PostgreSQL, pgvector, RAG, GraphRAG and LLM-agent ecosystem. Instead of treating a knowledge graph or a vector store as a place to hold present knowledge, Utopia puts time awareness and ontology in the base layer, so the knowledge system evolves as material arrives. Conflict detection, reasoning and decision making all run against that ontology, and the whole system deploys offline.

The concrete problem it solves is the loss of reasoning history. Where existing vector stores and knowledge graphs work to get present knowledge right, Utopia records the whole course of changing understanding. When a decision is reviewed later, the system can produce the full course it took and the grounds it rested on, which gives an organisation a compliance audit trail and a decision core its agents can trust. The project's authors explicitly describe this as a different route to enterprise intelligence, built bottom up from knowledge governance to trustworthy decisions and simulation, rather than an open-source take on any existing commercial platform.

Key capabilities

• A bitemporal knowledge graph, so knowledge carries both when it was true and when it was recorded.
• One Rust binary and one Postgres: full-text search is embedded in the binary, vectors go in pgvector, and the job queue is a table, leaving nothing else to run.
• A single web UI containing a system console, a graph browser and an ontology workbench.
• Conflict detection, reasoning and decision making evaluated against the ontology rather than against raw embeddings.
• Offline deployment, so a company can stand up a knowledge foundation and a compliance audit trail on hardware it controls.
• Agent-facing knowledge ingestion and recall, reflected in the topics agent-memory, rag, graphrag and llm.
• A published container image at ghcr.io/deeplethe/utopia on GHCR.

Who uses it and how

• Organisations that must keep enterprise knowledge and audit trails on hardware they own, since the system is designed for offline deployment.
• Teams building LLM agents that need a governed decision core instead of an unmanaged vector store.
• Compliance and review workflows where a past decision must be traced back to the grounds it rested on.
• Knowledge engineering efforts in enterprise records, education, finance, law and research, the public corpora the project states it has iterated against.
• Operators who want a small surface area: one binary plus PostgreSQL with pgvector, with no separate search cluster or queue service to run.

Getting started

The project ships a container image at ghcr.io/deeplethe/utopia, and the README provides a Quick start section alongside the official site at utopia.bi. Deployment expects PostgreSQL with the pgvector extension available to the binary.

How it compares

No list of paid products that this project replaces is provided in the facts, and no comparable tool is named in this registry, so Utopia stands alone here. The README does address one comparison directly, stating that the project would rather not be framed as an open-source take on Palantir, positioning it instead as a bottom-up route from knowledge governance to trustworthy decisions and simulation.

When to use it — and when not to

A self-hoster must operate PostgreSQL with pgvector and manage the deployment, since the offline, self-hosted model is the stated design and no hosted option appears in the facts. Teams that want a fully managed service, or that have no appetite for running a database, should look elsewhere. The entry also has an honest documentation caveat: the README excerpt available here is truncated mid-sentence, so detail on the feature set rests largely on that excerpt and the topic list, and the facts record no release history beyond the last push date.