Questions we hear most often

Ivertiq provides a sovereign Full-Stack AI Harness for regulated verticals: best-fit local models, Harness Control (policies, provenance, human-in-the-loop), and vertical workflow agents — deployed across a range from cost-competitive on-prem Node and high-throughput Station to siloed Private Cloud.

General tools optimize for horizontal productivity and often depend on public LLM APIs. Ivertiq is built for environments that require sovereignty, auditability, deterministic gates, and human accountability — joint optimization of models, control, and workflow apps inside customer-approved environments, not a chat UI bolted onto a cloud API.

Sensitive corporate data and operational systems are designed to stay in customer-approved environments (on-prem or private cloud). Core regulated workloads are not required to call public LLM APIs. Air-gap-capable deployments are supported where the site requires it.

Three jointly optimized layers:

1. Ivertiq Core Foundation — sovereign verticalized LMs, knowledge engine, inference optimization
2. AI Harness Control — orchestration, validation, provenance, human-in-the-loop (HITL), and determinism at the gates
3. Vertical Apps / agents — ERP ops, life sciences, and other regulated workflows

Harness Control is the operating layer; vertical agents are the apps. See how this maps to the industry AI software stack on the Platform page.

Ivertiq delivers sovereign intelligence across three dedicated commercial branches, all powered by the same Full-Stack AI Harness substrate:

• Branch 1 — Sovereign Enterprise Ops (SEOI): Non-invasive intelligence overlay on ERP across AP purchases, expenses, subcontracts, and AR export reconciliation under two-person segregation of duties.
• Branch 2 — Executive Assistant (Ivertiq Envoy): Vault-grade cognitive assistant for business executives and enterprise teams to whom sovereignty, data privacy, and intellectual property custody are non-negotiable.
• Branch 3 — Regulated Sciences: High-throughput forensic toxicology (mass-spectra LC–MS review) and GxP regulatory documentation (CTD, SOP compliance).

Post-training is part of the product, not a one-off science project. We shape open, best-fit local models for system capability — so the agentic layer can complete vertical workflows under customer SOPs — not for chatbot scorecards alone.

In practice that means more vertically specialized judgment and stronger agentic behavior: tool use, verification discipline, and efficient chain-of-thought. Quality and latency model tiers can be routed inside the Harness. Measured gains are workload-specific; illustrative ERP results and method detail are on the Technology page. Full eval packages are shared with design partners and investors under NDA.

Sovereignty is not useful if the workload will not fit the room. Inference-compute optimization is how Ivertiq makes useful agent concurrency run on the customer’s appliance — Node and Station on platform classes such as GB10-class and RTX 6000 / RTX PRO 6000-class — without metering every token to a public LLM API.

It is a product layer, not a prompt overlay: right-sized models and serving so production inference matches on-prem economics, while post-training can still run where GPU is abundant (private cloud). Technique choices are scoped per deployment; the buyer outcome is operable local serving, not a cloud-default model squeezed onto a box. See Technology and Deployment.

No. Owning weights is necessary and not sufficient. A closed-model API has a high floor and a low ceiling — you cannot improve the underlying intelligence with your Cases, SOP language, and reviewer corrections. A local general model without vertical-specific fine-tuning is still a general assistant in the customer’s room. Ivertiq is the third product: vertical-specific fine-tuning, system post-training, and a HITL Harness, served on-prem. The Harness stays; the local engine can be replaced. See Why Ivertiq.

Rather than using off-the-shelf general chatbots, Ivertiq deploys three standardized, vertically post-trained model engines optimized for specific enterprise operational roles:

• Ivertiq Swift (MoE A3B-class, ~35B total / ~3B active): Low-latency ingestion workhorse for high-speed document classification, invoice splitting, and barcode/QR extraction. Deployed on Ivertiq Node and Station.
• Ivertiq Prime (Dense 27B-class, ~27B total / 27B active): Enterprise reasoning core engineered for schema-constrained extraction, arithmetic verification, multi-way matching, and privileged database traversal. Deployed on Ivertiq Station (equipped with NVIDIA RTX PRO 5000/6000).
• Ivertiq Ultra (MoE 200B+-class, 200B+ total / ~20–30B active): Frontier sovereign reasoning engine for multi-entity tax consolidation, cross-border audit reconciliation, and regulatory dossier synthesis. Deployed on Ivertiq Private Cloud VPCs.

See technical specifications on the Technology page.

Governed path from document to decision: Ingest → Extract → Align → Govern → Assure → Brief & act.

Uncertain or consequential steps are human-gated. Systems of record (ERP, QMS, lab systems) stay authoritative. See the full path on the Platform page.

No. Ivertiq is a Harness that turns best-fit and vertically post-trained local models into controlled, auditable workflow execution inside the customer environment. Not a chatbot wrapper. Not a rented API.

For Sovereign Enterprise Operational Intelligence, we are an intelligence layer on the ERP they already run — we do not replace or re-platform the ERP. For Life Sciences, we implement existing SOP workflows under HITL; LIMS and QMS stay the system of record.

The human gate is part of the product. Nothing writes until a named person confirms. AI assists the reviewer; it does not replace them. Pilots can start from export/snapshot data before a write API exists.

Not silently. Consequential actions require human confirmation and, for ERP write-back, a controlled connector. Pilots may start with export/snapshot data before a write API exists.

Across a standardized deployment range mapped directly to our model engines:

• Ivertiq Node (GB10, Dual GB10): Cost-competitive on-premise appliance sized for edge/branch offices and individual accounting desks; hosts Ivertiq Swift (or quantized Prime).
• Ivertiq Station (NVIDIA RTX PRO 5000, RTX PRO 6000 Ada): High-throughput on-premise rack/tower for multi-user enterprise concurrency and executive suites; hosts Ivertiq Prime (and multi-stream Swift).
• Ivertiq Private Cloud: Highly scalable, multi-tenant siloed private cloud/VPC for multi-subsidiary enterprise fleets; hosts Ivertiq Ultra.

Same Harness stack; sized to economics, concurrency, and operating model. Deployment page →

We focus on regulated verticals where sovereignty, auditability, and deterministic validation are mandatory. Our three primary commercial branches are:

• Enterprise Operations (SEOI): Non-invasive intelligence overlay on ERP across AP purchases, expenses, subcontracts, and AR export reconciliation.
• Executive Leadership (Ivertiq Envoy): Vault-grade cognitive assistant for business executives and enterprise teams to whom sovereignty, data privacy, and intellectual property custody are non-negotiable.
• Regulated Sciences: High-throughput forensic toxicology (LC–MS spectral verification) and GxP regulatory documentation (CTD, SOPs).

We are defined by problem type (complex workflows, high-stakes decisions, regulatory oversight, explainable AI). Later domains follow productization of the first wedges — they are not the current agent catalog lead.

No. Beachheads may start in specific markets (e.g. Taiwan ERP reseller motion; life-sciences design partners), but the guiding constraint is regulated verticals, not geography. The product and deployment model are built for global expansion.

Start a pilot conversation via Contact. Typical first step: a scoped design-partner pilot with a named scope and human-in-the-loop gates. For ERP, a first look can start from an export or snapshot. Write-through waits for a named person.

No. On accounts a partner introduces, the partner owns the customer relationship. Ivertiq does not bypass the channel.

Partners and customers engage Ivertiq in two ways. Both run on the same sovereign Full-Stack AI Harness and the same deployment range (Node, Station, Private Cloud). What changes is packaging and who owns day-to-day customer engagement:

• Design-partner depth — for complex regulated workflows where Validation Pack fit, auditability, and specialist networks (e.g. CRO, MAH, forensic) decide the buy. Typical path: scoped pilot, then an owned or siloed deployment.
• Productized attach — for repeatable agents that attach to systems partners already sell — for example an intelligence layer on top of existing ERP for SMEs (we do not replace or re-platform the ERP). Partners own the customer relationship; we supply a Harness package sized for cost and concurrency, with human-in-the-loop on the high-frequency path.

Details on the Partners page.

An intelligence overlay on the ERP systems they already distribute, not a competing ERP. Resellers deliver our SEOI Case Desk covering high-frequency AP and AR accounting workflows under two-person Segregation of Duties (SoD).

Partners retain full customer relationship and renewal ownership; Ivertiq provides the appliance packaging, model engines, and deep-tech support. Write-through to the books always requires human authorization.

You meet customer demand for AI without expensive one-off ERP customization for every micro-difference. Small process quirks can often be absorbed in the AI/Validation layer. You keep renewal ownership; we provide product, demo kit, and deep-tech support.

Yes. A first conversation can use an export or snapshot. Write-through waits for a named person. Do not demo a posting that nobody confirmed.

Typically: 3–5 named account candidates willing to explore a pilot; a demo of payable and receivable cases; one agreed pilot scope; and, if useful, a start from an ERP export or snapshot. Write-through waits for a named person.

Adjacent motion: agents can help flag process-control breaks and assemble evidence packs. You own certification projects; we own continuous enforcement tooling — we do not sell ISO certificates.

Co-sell sovereign assistants under partner delivery models that respect sponsor, principal, and evidence confidentiality. Human review remains mandatory for accountable decisions. For forensic drug analysis, that often means assisting expert review of LC–MS / LC–MS/MS instrument PDFs — where database HIT scores are starting points, not sign-off. See Life Sciences and the forensic case study.

Common path: scoped pilot, then subscription and/or appliance (Node / Station / Private Cloud) packaging. Partner margin and territory terms are set in the partner agreement — directional economics are discussed commercially, not as website price lists.

Yes — enablement is part of the partner motion (demo script, positioning, do-not-demo boundaries, and technical follow-up). Deep architecture is for SE sessions, not the first sales call.