Connect the sources
ERP and CRM records, documents, conversations, images and field data arrive in different formats and at different rates. We retain raw sources while designing file, API and event-based ingestion around the work.
Data & AI
Enterprise data and
AI infrastructure.
Running a model is not enough for large, complex datasets. The system must preserve meaning, relationships and access boundaries while connecting user direction to operational workflows.
Xong / Connected operating model
Your data. Your control.SourcesUnified dataHuman + AIOperations
Permissions · provenance · execution history
See your data in a shared context.
Enterprise applications, documents, conversations and field data. We connect sources of different volumes and structures while preserving provenance and access boundaries.
- ERP / CRM / BPM
- Documents & media
- Sensors & events
Turn complexity into manageable knowledge.
We structure large, complex datasets from multiple sources, normalize fields, reconcile shared records and analyze correlations and relationships across sources.
- Normalization
- Entity resolution
- Relationship analysis
From intent to analysis. From analysis to action.
Dynamic AI agents assemble relevant data and tools around the user’s objective. Permissions, scope and approval rules determine which operations can proceed automatically.
- User direction
- Models & agents
- Review & approval
The whole operation, in one working environment.
We connect analysis, tasks, documents and decisions to enterprise process software. Teams access data, see work in progress and manage automated or semi-automated workflows from the same place.
- Shared records
- Process management
- Traceable outcomes
Data engineering
Data preparation and integration.
Establish shared meaning
Fields, dates, units and identities are normalized. Records referring to the same entity are reconciled across sources, with uncertain matches reviewed separately. Original and standardized values remain connected.
Make relationships visible
Documents are considered alongside the organizations, people, products or work records they concern. Correlations are assessed with source and time context; they are not treated as proof of causation.
Put information within reach
Full-text search, semantic retrieval and record relationships answer different questions. Users can reach information, inspect its source and understand how a result was produced. Access boundaries apply to retrieved results too.
The data lifecycle
Data records and traceability.
Raw records and analytical results are not the same thing. Each layer carries a different responsibility. This separation makes it possible to inspect the source, matching decision and processing step independently when a result is wrong.
Source record
Original file, system identity, version and ingestion time. Updating a source does not erase the version used by an earlier review.
Standardized record
Shared fields, units and date formats. A normalized value links back to the original value and the transformation rule.
Relationships & matches
Which organization, document or job does a record concern? Matching evidence and uncertainty are retained; an unverified relationship is not treated as fact.
Analysis & action
The user’s question, sources, recommendation and any approved system action. Retrieving information and changing a record are separate stages.
Working environmentInterface illustration · sample content
Scoped sources and access
Data ingestion
Original record and version
Original record and version
Collection
Source file
Extracted fields
Version record
Source and access scope
Source files and extracted fields remain distinct. When a version changes, previous results retain their source references.
Record relationships
Shared identity, sourced match
Shared identity, sourced match
Information fieldControl
Source recordSource format
Candidate matchShared representation
Review requiredTransformation rule
Source and access scopeHuman review
Records for the same work or organization are connected. Uncertain matches are not presented as verified facts.
User direction
From question to sourced analysis
From question to sourced analysis
Question and scope
Sourced findings
User assessment
Source and access scopeContext retained
Analysis across permitted sources is presented with evidence. Reading data and writing to external systems require different permissions.
User-directed AI
From user objectives
to governed agent execution.
Agents organize data access and tool use around a user objective. Business impact and permissions define the scope of automation.
User objective
The user states what they want to understand or do. Data scope, time range and permitted tools define the working boundary.
Dynamic planning & tools
The agent retrieves sources and assembles comparison and analysis steps. Access to a tool does not imply permission to execute every operation it supports.
Review & execution
Recommendations, evidence and proposed actions are assessed together. Work requiring human judgment goes to review; the target system’s outcome is recorded separately.
Source → recommendation → human decision → target-system outcome. Each is a distinct record; a recommendation is not presented as a completed action.
Platform modules
Platform components.
Explore how each module works and its technical scope on its own page.
An operable system
Evaluation and operations.
Resume after interruption
Long-running work needs recorded progress, error paths and retry behavior. Status should be visible to users; failure must not silently become success.
Evaluate on the same tasks
Model and prompt changes are compared on validated examples. Error types, data quality, execution time and human corrections are considered together.
Organization-fit operations
Data location, model providers, retention and access policies are part of deployment. On-premises, private cloud and hybrid options follow operational requirements.
How components work together
Between the data layer and the work layer.
Storing a source, interpreting it with a model and executing an external action are separate responsibilities. Connecting them does not turn them into one opaque AI box.
Data foundation
Raw sources, versions, standardized records and relationships.
Processing and retrieval
Data-appropriate processing and source-linked results.
Work execution
Steps, tools, permissions, error paths and target-system responses.
Review and improvement
Human judgment, correction history and comparable evaluation.
Components and operating approach
From raw records to user-controlled analysis.
Select a topic to explore the relationships between information, responsibility and delivery.
01
Structuring
Process documents, speech, imagery and application data according to their type. Carry identity, time, units and versions with the record.
Original value09.09.2026DEMO-DOC / v2
Standard value2026-09-09Source link retained
- Structuring
- Structuring
- Relationships and normalization
- Relationships and normalization
- Directed analysis
- Directed analysis
02
Relationships and normalization
Reconcile representations of the same entity across sources. Uncertain matches go to review; correlation alone does not establish causality.
Original value09.09.2026DEMO-DOC / v2
Standard value2026-09-09Source link retained
- Structuring
- Structuring
- Relationships and normalization
- Relationships and normalization
- Directed analysis
- Directed analysis
03
Directed analysis
Users define data scope and analysis objectives. Agents use permitted tools and present results with sources and any required human approval.
Original value09.09.2026DEMO-DOC / v2
Standard value2026-09-09Source link retained
- Structuring
- Structuring
- Relationships and normalization
- Relationships and normalization
- Directed analysis
- Directed analysis
An operating-model illustration, not live operations or measured performance results.
Work with us
Discuss your project with us.
Let’s assess your operations, data landscape and priorities together.