Utopia puts history into self-hosted RAG

Utopia’s strongest idea is treating RAG as a governed knowledge system rather than a pile of chunks. The tradeoff is that temporal truth, ontology maintenance, and human review add complexity most prototypes deliberately avoid.

• –Bitemporal facts preserve both when information was valid and when the system learned or corrected it, enabling point-in-time answers and audits.
• –Tantivy full-text search and pgvector embeddings are fused through reciprocal rank fusion, balancing keyword precision with semantic retrieval.
• –Review queues, undoable entity merges, evidence-linked facts, and temporal Datalog give developers practical controls against extraction errors.
• –Apache-2.0 licensing, OpenAI-compatible endpoints, and support for local providers such as Ollama and vLLM make air-gapped deployment plausible.
• –At v0.1, forward-only migrations and roadmap items such as MCP, SSO, and additional connectors mean teams should evaluate it as a fast-moving foundation, not a turnkey platform.