Eve-Legal™ Fusion v5 is the canonical configuration of the Eve-Fusion™ compound architecture for legal workloads. Five cooperating reasoning models compose dynamically per request. The classifier routes. The legal reasoner carries the domain workload. The frontier slots add nuance and cross-validation. Llama 4 Scout reads the entire case file in a single thought. The supervisor sits one layer above and orchestrates the whole loop.
Classifier
Sub-200ms routing across workload type, modality, reasoning intensity, and context-length needs.
Why this model occupies this slot. Lightweight 3.8B-parameter classifier. The first responder of the stack — decides which downstream models compose for the matter at hand.
Legal reasoner
LoRA-fine-tuned on Eve-Genesis (Law Edition). The domain-specialized workhorse for the bulk of legal reasoning.
Why this model occupies this slot. Carries the majority of legal workloads at near-Phi-4 cost and latency. Trained to reason like a senior litigator across canonical jurisdictional rules and procedural taxonomies.
Frontier reasoning
Best-in-class long-form structured reasoning. Used when the matter is high-stakes and nuance matters.
Why this model occupies this slot. Frontier slot one. Strong at demand-letter narrative, complex damages reasoning, and matters requiring careful disambiguation of authority.
Frontier reasoning
Frontier general reasoning. Complementary to Opus for cross-validation and ensemble routing.
Why this model occupies this slot. Frontier slot two. Provider diversity — when a client or jurisdiction prohibits a specific provider, this slot absorbs the constraint without rebuilding the agent.
Longitudinal context
10M-token single-context reasoning across the entire case file in one thought. The moat.
Why this model occupies this slot. The single largest open-weight context window publicly available. Workflow patterns this enables — full case-file review, multi-deposition synthesis, citation grounding across years of correspondence — are one-shot queries for Scout.
Five is a representative configuration, not a fixed schema. When the matter calls for it, a frontier slot can be swapped for a different provider; when a client or jurisdiction prohibits a specific model laboratory, the agent is not rebuilt — the constrained slot is reconfigured.
When the workload is light, the frontier slots step out of the loop entirely and the legal reasoner handles the request alone at SRM cost and latency. When the workload is heavy — a Daubert pre-read against a 40,000-page medical record — the long-context slot carries the load and the frontier reasoning slots cross-validate.
Above the five cooperating models, Justine — the named Digital Employee — runs the supervisor + sub-agent pattern. The supervisor decomposes the matter into stage-specialized sub-agents — intake, medical, valuation, strategy — and routes each sub-agent through the F5 reasoning loop with the calibration that stage requires.
The sub-agents are not separately branded. There is one Digital Employee: Justine. The supervisor pattern is what makes Mass Tort scalable — one Justine supervising thousands of plaintiff sub-agents in a single reasoning context.
For PI principals, managing partners, and litigation operators evaluating reasoning-grade AI for their firm. Self-serve trial available for solo and small practices; sales-assisted for mid-size and enterprise.