praesensus
Architecture

System of insight, not system of record.

Your practice-management platform stores what happened. Praesensus tells you what it means. Five layers. One codebase. Three deployment postures. Every architectural choice is documented, measurable, and reproducible.

The fundamental distinction

Clio, Filevine, MyCase, Smokeball, iManage, NetDocs, Relativity — every legal-technology platform in market today is a system of record. It stores case data, tracks time, manages documents, automates forms. Each is a silo. Each sees only what you enter into it. AI features, where present, answer questions about data already inside that system.

Praesensus is a system of insight. It reads from every system of record the firm already uses — without replacing any of them — fuses the signals into a live operational picture, recommends the next action, governs it through a named-person approval ladder, and measures its own accuracy against real data.

The shift changes the operating posture of the practice. Instead of attorneys manually correlating information across platforms, the intelligence layer does it continuously. Instead of deadlines depending on someone's memory, they are projected from events. Instead of contradictions hiding until litigation, they surface on detection. Instead of governance depending on discipline, it is enforced by architecture. Instead of AI accuracy being a matter of faith, it is a matter of measurement.

The five layers

L5 · Verticals

Litigation · Contracts · Compliance · Personal Injury

Each vertical is a profile + taxonomy + monitor + skin. New verticals do not require core-code changes; the same spine underneath every one.

L4 · Read-only lens adapters

Ingest · Extract · Reason · Export

Every integration point is a plugin, not a fork. Standards-based (SAML/SCIM/LDAP) + legal-vendor-specific (Clio, Filevine, MyCase, Smokeball, iManage, NetDocs, Relativity, PACER, Westlaw). Data is read, not replicated.

L3 · Combining Layer + Knowledge Graph

18 node types · 16 edge types · genuinely new IP

Person · Organization · Matter · Contract · Clause · Obligation · Email · Document · Court · Judge · Statute · Regulation · Precedent · Filing · Deadline · Billing · Evidence · Approval. Identity resolution, state fusion, evidence fusion, timeline fusion, semantic linking, contradiction detection, cross-plugin correlation. Produces conclusions no single source could reach.

L2 · Legal State Spine

Seven-stage deterministic pipeline

OBSERVE → NORMALIZE → DETECT_STATE → DETECT_TRANSITION → ESTIMATE_OUTCOME → RECOMMEND → VALIDATE. Lifted intact from four production-tested Atlas siblings. Rule-based classifiers, stdlib-only Python, zero third-party runtime dependencies.

L1 · Legal data

Read-only adapters. One-way ingress.

Matters · Documents · Emails · Contracts · Court records · Dockets · Calendars · Billing · Depositions · Statutes · Regulation feeds · Precedent · Medical records · Firm knowledge stores. Nothing modified. Nothing exfiltrated. The privileged content stays in your firm.

The 7-stage Legal State Spine

Every advisory travels the same seven-stage pipeline. Deterministic. Reproducible. Explainable at every stage.

Stage 1

OBSERVE

Ingest events from every read-only lens adapter. Signals typed into canonical schema.

Stage 2

NORMALIZE

Extract clauses, obligations, entities, dates. Canonicalise across sources.

Stage 3

DETECT_STATE

Interpret each matter's current operational state — on-track, at-risk, breach-imminent.

Stage 4

DETECT_TRANSITION

Debounce state changes across persistence windows. Catch drift early without noise.

Stage 5

ESTIMATE_OUTCOME

Forecast deadlines-to-breach, exposure, cycle time. Trajectory calls, not point predictions.

Stage 6

RECOMMEND

Synthesise fused inputs into one action per decision. With named drivers.

Stage 7

VALIDATE

Deterministic replay + precision/recall metrics. Every advisory reproducible.

The dividing line

Perception vs. authorization

Stages 1-5 are perception. Stages 6-7 are authorization. The dividing line is where the fail-closed governance gate sits — default-deny with a named-person approval tier ladder.

22 capabilities across 4 domains

Every claim Praesensus makes about the intelligence layer breaks down into 22 specific, measurable capabilities across four architectural domains. Every one has a corresponding entry in the competitive comparison — and in every one the structural difference between a system of record and a system of insight produces a qualitative gap no amount of incremental feature development on the record-keeping side can close.

Domain 1

Intelligence & Analysis

Cross-matter reasoning, contradiction detection, deadline projection from cause not calendar, entity resolution across sources, portfolio-scale correlation, validated-accuracy metrics.

Domain 2

Architecture & Integration

Read-only lens adapters (Clio, Filevine, MyCase, Smokeball, iManage, NetDocs, Relativity, PACER, Westlaw), 18-node knowledge graph, 7-stage deterministic pipeline, zero third-party runtime dependencies.

Domain 3

Governance, Risk & Accountability

Default-deny fail-closed gate, named-person approval-token ladder, reason-coded HALT paths, SLA-triggered escalation up the roster hierarchy, ALCOA+ audit trail with unbounded retention.

Domain 4

Portfolio & Operational View

Live portfolio state across every matter, cross-firm density-unlocked benchmarks (at L3 federation gate), federated pattern intelligence, standard-form defect signals, regulator-behaviour signatures.

Full capability-by-capability comparison against Clio, Filevine, MyCase, and Smokeball available under NDA — request via contact.

Validated accuracy

Every competitor markets AI-powered features. None publishes accuracy metrics against public legal benchmarks. Praesensus does.

CUAD

72.9%

Contract clause understanding.

GDPR Enforcement

96.9%

Enforcement action classification.

CaseHold

75.6%

Legal reasoning benchmark.

Full audit: 386 tests pass in 2.71s · 57% test-to-source LOC ratio · 9 real-data proxy pilots across ~7,000 records. Deterministic replay harness ships with the product — every advisory reproducible from source data at any timestamp. Full validation report and pilot-runner scripts available under NDA.

Three deployment modes

Air-gapped first. Same codebase. Same behaviour. Deployment posture is a config flag, not a build target.

Air-gap on-prem

Zero automated outbound. Federation via signed daily pull + monthly push, mTLS + HSM authenticated. Zero third-party runtime dependencies means the entire product runs offline.

Best for: Healthcare payer, defence contractor, F100 regulated finance, government legal.

On-prem connected

Signed appliance in your datacenter. Federation via mTLS. Customer-scheduled updates. All data stays inside the firm perimeter.

Best for: Am Law 50-100, F500 GC offices.

Cloud tenant

Praesensus-hosted, per-firm namespace, customer-selected region. Continuous updates.

Best for: Mid-market in-house, SaaS-native firms, PE portfolios.

Four appliance forms

Physical

Pre-provisioned 1U/2U server. TPM sealed. Regulated verticals.

Virtual

Signed OVA/OVF/VMDK. VMware/Hyper-V/KVM. Default for BigLaw + F500.

Container

Signed Helm chart + OCI images. Kubernetes.

Cloud SaaS

Multi-tenant SaaS with per-firm namespace.

Integration surface

Praesensus integrates with the standards-based half of firm IT via native protocols (no custom shim required), and with the legal-vendor-specific half via read-only plugin adapters.

Standards-based (no custom code required)

CategoryCommon vendorsProtocol
Identity providerAzure AD, Okta, Ping, ADFSSAML 2.0, OIDC
Directory / rosterSame as aboveSCIM 2.0, LDAP, CSV bulk
SIEM / monitoringSplunk, QRadar, Sentinel, Datadogsyslog, Splunk HEC, webhook
PKI / certificatesFirm CA, DigiCert, EntrustX.509, PKCS#12
Backup / snapshotVeeam, Rubrik, Cohesity, CommvaultStandard VM snapshot APIs
Vulnerability scanRapid7, Qualys, TenableStandard VA agents inside appliance

Legal-vendor-specific plugin adapters

CategoryVendors
Email / CalendarMicrosoft 365, Google Workspace, on-prem Exchange
Document managementiManage, NetDocs, SharePoint, OpenText eDOCS
Matter / practice managementClio, Filevine, MyCase, Smokeball, Litify, Aderant Elite, Elite 3E
Contract lifecycle mgmtIronclad, Icertis, ContractPodAi, Agiloft, DocuSign CLM
eDiscoveryRelativity, Everlaw, DISCO, Reveal
Court + docketPACER, state e-filing, docket-alert services, CourtListener
Legal researchWestlaw, Lexis, Bloomberg Law, Fastcase
Billing / timeAderant Elite, Elite 3E, TimeSolv

Every adapter is read-only. Praesensus observes what your systems produce; it does not write back, replace, or replicate. The privileged content stays in your firm.

Federation architecture

The Federation is the anonymised pattern-intelligence network that enables cross-firm benchmarks and standard-form defect detection. It operates without ever accessing your privileged content.

Federate patterns, not privileged content. The Federation processes only derived typed structure and outcome patterns — clause frequency signatures, deadline pattern statistics, judge outcome distributions, standard-form defect signals. It never processes and never accepts privileged text, client identifiers, matter identifiers, or attorney communications. Enforced architecturally.

Asymmetric flow

  • Downstream (Praesensus → customer): daily authenticated pull. Fresh federated intelligence in.
  • Upstream (customer → Praesensus): monthly signed push. Human-inspectable pattern digest out.

Authentication

Two-factor machine authentication for automated pull: mTLS client certificate (90-day rotation) + hardware-backed signed request token (HSM/YubiKey PIV). Optional weekly or monthly human Custodian Seal Renewal.

Composed foundations

Praesensus is the fifth vertical instance of the Atlas platform family. Every core module — the seven-stage spine, the governance gate, the transition detector, the deadline forecaster, the validation harness, the runtime, and the plugin surface — is lifted intact from production-tested siblings serving buildings, biopharma manufacturing, genetics, and critical facility infrastructure.

The Combining Layer + Knowledge Graph is the genuinely new IP for legal. Composition beats invention.

Zero third-party runtime dependencies. No LLM API calls at runtime. No SaaS AI providers. No vendor lock-in for the reasoning layer. The entire product runs on stdlib-only Python with rule-based classifiers — which is why air-gap deployment is a config flag, not a special build.

Read the origin story →

Deep-dive with your technical team