Utopia: Open-Source Enterprise World Model Adds Bitemporal Graphs & MCP for AI Agents

01 RAG Finds Documents but Cannot Answer Historical Questions

RAG retrieves relevant passages for questions like "What is the refund policy?" but fails for temporal queries: which payment terms applied to a contract on March 15, 2024, or what customer status the system held in February. Utopia's README illustrates this with a Master Services Agreement that underwent multiple amendments; the system must return the terms valid on a specific date and cite the source amendment.

If the knowledge layer only stores the latest value, historical queries distort. Example: a customer_type updated from SMB to Enterprise in April. An audit six months later of a February pricing decision would see the wrong state. Agents participating in contracts, risk control, and operations need to know what facts the system believed at that past time .

02 Bitemporal Graph: Two Timelines for Every Fact

Utopia's core structure is a Bitemporal Graph . Each fact carries two timelines:

Valid time – when the fact was true in the real world.

Transaction time – when the information system recorded or modified it.

Example: a contract effective March 1, but ERP synced March 5; an amount error discovered in April, corrected in May. Utopia does not overwrite; it ends the old fact's validity interval and creates a new version with replacement links . An agent receives not just contract_value = X but also when it was valid, when it entered the system, what superseded it, and its source.

This preserves the evolution of business state, a task previously scattered across audit logs, version tables, and manual processes.

03 Conflict Resolution and Human Review Built Into the Pipeline

Enterprise data arrives from CRM, ERP, contract platforms, data warehouses, emails, and meetings — often describing the same entity with conflicting states. Utopia integrates:

Ontology -driven entity and fact extraction.

Entity Resolution via name/alias matching, then Embedding Similarity, then model judgment for uncertain cases.

Conflict Detection for cardinality violations and contradictory facts.

Review Queue for low-confidence extractions, duplicate suspects, and constraint violations — routed to humans.

Conflicts are not auto-resolved by "new wins". Options include: close old fact and accept new, keep both versions, or reject new fact. Ontology constraint conflicts allow retracting a fact, modifying the rule, or accepting the conflict.

Reasoning uses forward chaining on transitive, symmetric, and inverse relations defined in the ontology, but derivation is disabled by default because erroneous rules propagate errors. Derived facts are marked separately with provenance; explicit facts override derived ones .

Result: the knowledge base becomes a versioned, evidenced, revisable business state rather than a static answer.

04 MCP Integration: Agents Query State, Write Proposals

Utopia treats agents as first-class consumers. It provides a built-in agent and an MCP Server for the knowledge base. MCP clients (Claude Desktop, Cursor) query with permissioned tokens. The design principle: "Reads are typed and dated, writes are proposals."

Agent reads typed, time-stamped facts; agent writes enter as proposals for human confirmation.

Contrast with RAG: RAG dumps text chunks, leaving entity resolution, temporal logic, and conflict handling to the model's context window. Utopia pre-organizes the business state — entity identity, fact type, validity interval, provenance, replacement links — so the agent receives a structured state, not raw text.

A Decision Ledger (append-only) records every confirmation, rejection, merge, revocation, and graph rebuild with actor, timestamp, and target object. History is never overwritten by current state. When agents drive contracts, procurement, or risk control, teams can audit: why was this fact believed, what was the source, who confirmed the change, and was the decision-time state identical to today's view?

05 "Enterprise World Model" Is a Direction, Not a Mature Category

The term evokes RL world models (Ha & Schmidhuber, 2018) where agents learn compressed environment dynamics for planning. Utopia differs: it does not simulate enterprise operations or predict state evolution . It builds a queryable, updatable, time-versioned, provenance-tracked business state layer — customers, contracts, organizations, relationships, fact validity, system knowledge time, and modification history.

Roadmap items (Decision Intelligence, decision replay, Execution Gate, Business Rules, Agent Memory over MCP) remain in development. Maturity: v0.1 ; schema may change, migrations are forward-only with no rollback; production use requires version pinning and backups.

The shift: enterprise agent context moves from a static retrieval corpus to a continuously maintained business state . The open question is whether this state can stay accurate and maintainable without ontology, review, and versioning becoming new engineering burdens.

Source: DeepLethe / Utopia GitHub Repository & README