CONTENT AND PROCESS ENGINE

Everything is typed

Records have schemas. Process steps have schemas. Both validate data at the boundary, not at runtime through guesswork.

Declared writes

Every step declares exactly what it is allowed to change. An agent reviewing a document cannot take an action the process does not permit. The engine owns that rule, not the AI.

Permission-aware everywhere

Reads, writes, and process transitions all check permissions at the moment of action, based on the content, the user, and the policy in force. Nothing crosses a line the user is not entitled to cross.

Versioned and audited end-to-end

Documents, schemas, processes, and outputs all carry a version and a full history. Every approval and change is queryable in one structured trace.

Smarter than search. Faster than people.

Old-school content systems let you search for a word inside a document. Vertesia lets you ask a question and get a real answer instantly, across every content store you own.

• Semantic search: Ask in plain English. Get direct answers instead of a pile of files to sort through.
  
  
• Automatic enrichment: AI extracts insights, relationships, and key data from every document on ingestion.
  
  
• Cross-repository intelligence: Works across SharePoint, file shares, legacy ECM systems, and more, all at the same time.

Workflows that do the work, not just route it.

A traditional workflow sends a document to a person. Vertesia's Process Engine sends the document to AI which reads it, makes decisions, and completes tasks. Humans are only pulled in when a judgment call is needed.

• AI agents: Handle multi-step tasks that require reading, reasoning, and decision-making.
  
  
• Automated tools: Execute fast, predictable steps like file conversion, record matching, or system updates.
  
  
• Human checkpoints: Bring in a person only when truly needed, with full context already prepared for them.
  
  
• Enforced guardrails: Every process transition is gated by code, not AI. Rules cannot be argued with or bypassed.

Content objects, types, and collections

Any type of content (documents, images, audio, video, and structured data) is stored as a typed object. Content types define what fields are required, what values are valid, and how records relate to each other. Collections group content by project, team, or business unit.

Metadata and lifecycle

Key information is pulled from every document automatically on arrival and stored as structured, searchable data. Each document moves through defined lifecycle stages and retention rules are enforced automatically.

Versions and renditions

Every change is tracked. You can compare versions, branch for translations or redactions, and merge changes back. Preview images, text layers, and formatted exports stay attached to the source file.

Permissions

Access is controlled at the document level, not just the folder level. Who can see a document depends on its status, the user's role, and the active policy, all evaluated in real time at the moment of retrieval.

Search

Advanced search is made possible with Elasticsearch, offering exact phrase, filter by structured metadata, or ask a question in plain English. Agents and users can use all three search types in a single query.

Lineage and audit

Every derived file records where it came from and how it was made. Every access is logged, attributed, and queryable. The audit trail is comprehensive and accurate, not rebuilt from scattered log files.

Typed context and declared writes

Every step in a process declares exactly what information it is allowed to add or change. The engine enforces this strictly so no agent can take an action outside what the process permits.

Explicit transitions

A process moves forward based on clear, defined conditions, not on what an AI happens to say. Transitions are triggered by data conditions, agent decisions from a fixed set of options, or human signals.

Human tasks

Long-running processes can pause and wait for a human to review -- without wasting compute time. Human review is a first-class feature, not an edge case.

Supervisor intervention

An optional supervisor agent can guide a process through approved commands without rewriting the process definition. Every command is logged with a full audit record.

BPMN import

Already have process definitions or diagrams? Import them. The engine converts supported patterns automatically, preserves the original, and clearly flags anything it cannot convert.

Authored by AI, run by the engine

A built-in assistant can draft a process from a plain text description, a screenshot, or a whiteboard photo. Every definition is validated before it runs. The assistant can simulate, review, and suggest revisions for human approval.