The vocabulary settling into the category.

AI systems that act with agency on behalf of the people and institutions they serve, rather than passively answering questions. Agentic systems decompose work, coordinate sub-tasks, take action under supervision, and produce work product that is reviewable at every step. The opposite category — passive question-answering — is not agentic.

A vertical Agentic AI product calibrated for a single industry. JustineAI™ is the Agentic Legal OS; ChironAI™ is the Agentic Healthcare OS; ArthurAI™ is the Agentic Learning OS; TheoAI™ is the Agentic Theology OS. Each is structured to coordinate the structured workflows of its industry — not to answer chat questions.

A named agent within an Agentic Operating System who coordinates stage-specialized sub-agents on behalf of the customer. Justine is the JustineAI™ Digital Employee. Digital Employees have persistent identity, supervised action, and named accountability. They are not chatbots.

The architectural posture where a Digital Employee supervises stage-specialized sub-agents that handle specific phases of the matter. The sub-agents are not separately branded. The pattern enables scale (one supervisor coordinating thousands of plaintiff sub-agents in Mass Tort), preserves attorney-client framing, and produces a structural audit trail.

See the supervisor-pattern essay.

A multi-model architecture where cooperating reasoning models compose dynamically per request, with no single load-bearing element and no fixed dependency chain. Contrast with a stack, which is a fixed dependency chain anchored to a single foundation model.

See the compositional-fabric essay.

The canonical configuration of the Eve-Fusion™ compound architecture: five cooperating reasoning models composed per request — Microsoft Phi-3 classifier, Microsoft Phi-4-derived Small Reasoning Model fine-tuned on Eve-Genesis (Law Edition), one to three commercially available frontier models, and a long-context model (Meta Llama 4 Scout). Eve-Legal F5/reasoner is the JustineAI™ instantiation.

A compact, domain-calibrated reasoning model — typically a fine-tuned Microsoft Phi-4 — that carries the bulk of domain reasoning load at near-Phi-4 cost and latency. The SRM is the workhorse of the compositional fabric; the frontier slots step in only when the matter justifies the spend.

The proprietary synthetic reasoning corpus that fine-tunes JustineAI™’s legal SRM. 100% synthetic by construction — no customer matter data, no scraped privileged communications. Coverage is engineered to the production workload; bias is documented; the corpus extends as new editions ship.

See the synthetic-data essay.

MindHYVE.ai™’s proprietary cloud architecture, anchored to Microsoft Azure. Tenant isolation, US data residency by default, inherited platform attestations (ISO 27001, ISO 27018, SOC 1/2/3, PCI DSS, HITRUST), and customer-managed keys on enterprise tiers.

A reasoning posture in which the AI accepts a context window large enough to read the entire matter — for PI, 5,000 to 50,000 pages — in a single inference call, without chunking or retrieval-augmented hand-off. Meta Llama 4 Scout at 10M tokens occupies this slot in Eve-Legal F5/reasoner.

See the ten-million-tokens essay.

The pre-long-context approach of breaking the matter into smaller chunks, embedding the chunks into a vector store, and retrieving the chunks the prompt most resembled at inference time. Long-context reasoning retires the constraint that made chunking necessary; chunking is preserved only for matters that exceed even the long-context window.

The four-pass demand-letter pipeline’s fourth-layer QA check. Every cited legal authority is resolved against the live CourtListener corpus at generation time. If a citation does not resolve, it is removed and the matter is flagged for the attorney of record.

The construction sequence for a JustineAI™ demand letter: draft, medical-causation pass, damages quantification pass, four-layer QA. Each pass operates on the output of the previous one. The pipeline’s output is attorney-attested before service.

The MindHYVE.ai, Inc. corporate structure where each Agentic Operating System is operated by its own wholly-owned LLC: Eve-Legal, LLC for JustineAI™; Eve-Healthcare, LLC for ChironAI™; Eve-Education, LLC for ArthurAI™; Eve-Theology, LLC for TheoAI™. The contracting counterparty for the JustineAI™ product line is Eve-Legal, LLC; the corporate parent is MindHYVE.ai, Inc.

The principle on which MindHYVE.ai was founded: the case-reasoning quality historically locked inside elite institutions can be made universally accessible through the right ethical deployment of agentic AI. For JustineAI™, this means any litigator with the product has access to the case-reasoning quality elite firms have long held as their structural advantage.