Precision
Every answer is traceable to doctrine. No fabrication, no approximation, no confident guessing. If the field manual says it, we say it. If it does not, we say that too. Precision is non-negotiable when the stakes are real.
Mission
Why Elysian Exists.
General-purpose AI is optimised for plausibility, not accuracy. When the question involves tourniquet placement, dead reckoning without GPS, or water procurement in an arid environment, a confident-sounding but fabricated answer is not an inconvenience — it is a liability. Every major AI assistant available today draws from the same broad pool of internet-scraped training data, with no reliable mechanism to distinguish verified military doctrine from forum speculation. For survival and tactical queries, that gap is unacceptable.
Elysian was built to close that gap. Every response is grounded in indexed, verified field manuals and military doctrine through retrieval-augmented generation. If the answer exists in the knowledge base, Elysian retrieves it, structures it, and cites the source. If it does not, Elysian tells you — clearly, without hedging or substitution. The system is designed to fail with honesty rather than succeed with fiction.
Technology
How the Knowledge Base Is Built.
From raw doctrine sources to doctrine-accurate answers — in three stages.
Source Ingestion
Curated military field manuals, survival doctrine, and TCCC guidelines are converted to clean structured markdown. No OCR noise, no formatting artefacts — only the doctrine.
Sources include FM 21-76, SAS Survival Handbook, TCCC Guidelines, ATP 3-50.3, FM 3-05.70, NATO STANAG 2928, USMC SERE Manual, and WEMS Wilderness First Aid protocols.
Chunking & Embedding
Doctrine is segmented into ~500-token chunks with contextual overlap and converted to vector embeddings. Each chunk retains its source reference and section context.
Vectors are stored in the cloud. At query time, your question is embedded with the same model and matched against the corpus — returning the top-k most semantically relevant passages.
Anthropic-Powered Synthesis
Retrieved passages are injected as grounded context into Anthropic Claude. The model synthesises a structured, actionable answer — constrained entirely to the retrieved doctrine.
Claude is instructed to cite sources, distinguish between phases of care or tactical scenarios, and refuse to answer outside the retrieved context. It does not draw on its general training data for field-specific responses.
Core Values
What We Stand For.
Reliability
Elysian is built for moments when information has to be right the first time. That standard governs every decision about how the knowledge base is constructed, maintained, and extended — conservatively, not aggressively.
Accessibility
Military doctrine is dense by design. Elysian translates — converting technical field manual language into answers a trained person can act on immediately, without requiring a library card or a prior service record.
Mission Focus
We do not optimise for engagement metrics, session length, or virality. We optimise for accuracy under pressure. Every product decision is evaluated against one question: does this help the user make a better decision in the field?