Enterprise AI projects often fail not because the models are inadequate, but because organizations lack a structured substrate that encodes how decisions are made, negotiated and executed. Generic LLMs carry no firm‑specific ontological priors; retrieval‑augmented generation is brittle and provides no path to executable actions; static playbooks capture logic but cannot reason or adapt. An architecture that co‑designs tacit‑knowledge capture, ontological grounding, sovereign deployment and auditable actuation from the outset is therefore required.
This paper introduces the Corporate Language Model (CLM), a framework that transforms a firm’s structured, unstructured, multimodal and tacit knowledge into an ontology‑grounded enterprise foundation on which reasoning and governed execution are composed. CLM consists of five capability planes and four architectural pillars: the Neurosymbolic Mesh couples generative models with a knowledge graph; the Skill Graph types reusable tactics, personas, objections and goals and enables composition; Living Digital Twins model functional areas as reasoning surrogates; the Deep Security Layer enforces sovereignty, traceability and human oversight; and the Spec‑as‑Code paradigm bridges grounded intent to executable artifacts.
Four contributions are presented: (1) CLM is defined as a distinct object of study; (2) the Skill Graph is introduced to achieve compositional explainability by construction; (3) the Wisdom Listener effect is proposed, showing that tacit‑capable foundations compound in value with use, linking to dynamic capabilities and organizational learning; (4) evidence from a JCI‑accredited tertiary hospital in Brazil demonstrates three of the six LGPD‑aligned maturity stages.
Review