Primo AI Server

Document AI for robots and agents —
without per-page pricing

Primo AI Server processes structured and unstructured documents — invoices, contracts, claims, statements, foreign trade paperwork — and turns them into structured data for RPA robots and Primo Epic agents. Deployed on-premises. Licensed by capacity, not by page volume.

See the platform

RPA + AI

RPA executes rule-based tasks with full repeatability.

Form submissions, data transfers, system integrations — when the logic is explicit, RPA runs it consistently, with a full audit trail and no variance.

AI understands documents and free text where rules cannot be written.

Document fields that shift across suppliers, free-text inputs, classification by content rather than filename — these require understanding, not a rule set.

AI Server adds document understanding, on the customer’s own infrastructure.

Capabilities

Four AI capabilities. Documents are the core

Arabic OCR and intelligent document processing

Extract structured fields from Arabic and English documents — invoices, identity documents, forms, customs declarations. This is intelligent document processing in its classic scope: known document types, defined fields, high volume.

—Train on around 50 labeled samples in the UI — no ML team or separate training infrastructure needed
—Unstructured documents — full text extraction without training

Document classification

Classify documents before extraction to route each one to the right pipeline. Supports named document types and mixed batches. Configurable in the UI — no separate ML infrastructure.

AI Text — LLM document processing

Apply extraction, classification, summarization, and generation over raw text or directly on document scans via multimodal LLM.

—Connects to any local model or external LLM API
—Switch models without rebuilding pipelines
—Works on scanned PDFs and images via multimodal inference

Document AI for Primo Epic agents

When a Primo Epic agent receives a scan or an unstructured document, AI Server recognizes it and returns structured fields. The agent then checks them against company rules. The same AI Server serves robots and agents.

Several design decisions set AI Server apart from cloud AI services and standalone IDP software — in pricing model, deployment topology, and how tightly document AI integrates with the automation layer.

Differentiators

Sovereign AI: what sets AI Server apart

No per-page pricing

Pay by capacity, not document volume.

No ML team required

Analysts and RPA developers configure and train.

No vendor AI lock-in

Connect supported local or external LLMs through approved APIs. Switch models without rebuilding workflows.

One document layer for robots and agents

RPA robots and Primo Epic agents use the same recognition and extraction models.

Classify, then extract

Mixed batches are sorted by document type before fields are extracted.

Air-gapped deployment

See deployment architecture →

For teams with an existing stack, AI Server typically consolidates tools currently managed as separate products.

What AI Server replaces in your stack

Without AI Server

—Separate IDP platform with its own licence, integration, and per-page billing
—Cloud-only OCR/NLP services that send documents off-premise
—Custom ML team to train and maintain extraction models

With AI Server

—Document AI integrated into the same platform, the same orchestrator, the same audit trail
—On-premises execution, including in air-gapped environments
—Configuration by analysts and RPA developers, no ML team needed

Not every deployment is a fit. Here is an honest read on where AI Server performs best — and where it may not be the right call yet.

When AI Server is the right choice

Good fit

—High document volume with consistent document types
—Existing RPA program needing to handle unstructured input
—Air-gap or data-residency requirements
—No ML team, but a need for production AI
—Predictable cost matters
—Regulated document sets that cannot leave the premises — including hospital and payer document workflows

Watch out for

—No clear success criteria defined before starting
—Primarily handwritten or very low-quality source documents
—No budget for GPU infrastructure
—Highly variable input with no available labeling data

What AI Server delivers
at document scale.

−50—70%

Manual document handling at scale

Typically, across deployments handling invoices, contracts, and foreign trade paperwork.

−40%

Lower IDP cost vs. separate-platform alternatives

Capacity-based licensing eliminates per-page fees at volume.

Documents leave customer infrastructure

In on-premises and air-gapped deployments with external services disabled.

Ranges synthesized from industry analyst reports and Primo's deployments across document-intensive environments. Individual deployment results depend on baseline maturity, process scope, and integration complexity.

Get started

See AI Server on your own document workflow