# Humans and AI understand design on the same page.

LAPIS3D connects P&IDs, 3D models, document clauses, requirements, and change history so AI agents can understand real engineering objects and review design.

**Engineering Context Infrastructure for Industrial AX**

## The problems EPC and Owner Operators actually face

- **Keeping massive models current for daily review is hard** Smart3D mega-models take so long to fully re-extract and redistribute that sharing the latest model lags. S3DX incrementally updates only what changed, so the latest 3D model can be deployed quickly. HQ, site, and partners can share and review design against a fresh model every day.
- **Repetitive design review consumes a lot of engineer time** Drawings, 3D models, document clauses, and requirements sit apart, so engineers must find targets and confirm relationships by hand. LAPIS3D lets an AI Agent understand Engineering Context, linking document clauses to real objects and space to support repetitive design review and impact analysis.
- **You can see revision differences, but not what they affect** Knowing only what changed in a new revision is not enough to judge design impact. LAPIS3D connects changed objects to tags, P&ID relationships, and surrounding 3D space to analyze the impact scope of Revision Deviation. Engineers can see both the before/after difference and what needs review.
- **The basis for design-review judgments disappears easily** Screenshots and comments alone make it hard to revisit which model revision, objects, document clauses, and space informed a judgment. Evidence Snapshot records the Viewpoint, objects, revision, requirements, and judgment basis from the moment of review. You can trace and share not only the result but why it was judged that way.
- **P&ID tags and real 3D model objects do not connect automatically** Even for the same equipment, P&ID and 3D model names, hierarchies, and data structures can differ. Tag Resolver uses Engineering Identity and model context to find the real 3D object a tag points to. People and AI Agents analyze and review design against the same target.
- **As AI use grows, so does the On-prem GPU infrastructure burden** EPCs and Owner Operators mostly run design data On-prem. nanochat uses Engineering Context and GraphRAG to reduce large-LLM load and run local AI Agents on existing CAD workstations, lowering added GPU infrastructure dependence and cost.

> LAPIS3D solves these problems with Engineering Context.

[Explore industry workflows →](https://lapis3d.com/en/industries/)

## Process Context. Spatial Context. One engineering reality.
Process Rich Tag Graph connects engineering identity, process semantics, and records. Spatial Graph connects objects, hierarchy, spatial relationships, and geometry. Shared identity and provenance keep them on the same page.

**Process Rich Tag Graph** — We collect data from the SmartPlant P&ID Database or key engineering lists, normalize objects, attributes, identifiers, and relationships against the DEXPI 2.0 specification, and build a Process Rich Tag Graph. Different sources can then be interpreted through the same Engineering Identity and Process Context.

- Engineering identity
- Process semantics and topology
- Records and provenance

**Spatial Graph** — The Spatial Graph is built from Smart3D objects, hierarchy, position, relationships, and metadata — and, where needed, from other key 3D engineering models.

- Object identity
- Model hierarchy
- Containment, adjacency, support
- Geometry and provenance

Humans read this through P&IDs, drawings, 3D, timelines, and Evidence Snapshots. AI reads the same identities, graphs, metadata, and provenance.

## Find the object. Then find what matters around it.
Tag Resolver resolves engineering identity to the authoritative target object. Spatial Resolver interprets engineering intent through Spatial Graph and qualified spatial rules to find the relevant neighborhood and relationships.

**Tag Resolver finds the object.**
Authoritative tag context, 3D naming patterns or regex, hierarchy, and model context resolve toward the target 3D object. Where ambiguity remains, it returns confidence, a candidate set, provenance, and a human review path.

**Spatial Resolver finds the context.**
Engineering intent, the identified object, Spatial Graph, and qualified spatial rules produce the neighborhood an engineer actually needs to review. It is not proximity search, nearest-neighbor search, or scene-graph traversal.

A qualified spatial rule carries its source, intent, applicability, target class, relationship type, parameters, validation state, and provenance. The result can explain which rule produced the neighborhood and why.

> From document retrieval to object resolution. From object resolution to spatial reasoning.

## AI should not just answer. It should show what it saw.
An Evidence Snapshot is a portable engineering context package, not a screenshot. It links the view, object identity, tag, revision, requirement or clause, markup, review context, provenance, and spatial rule attribution where applicable.

> Evidence Snapshot makes the decision legible.

- **View:** 3D viewpoint, clipping, section
- **Object:** Object identity + tag
- **Revision:** Model / drawing revision
- **Requirement:** Requirement or clause
- **Markup:** Comment, finding, annotation
- **Review:** Review context and disposition
- **Provenance:** Source, rule, version, time
- **Rule attribution:** Which spatial rule produced this neighborhood

Portable is a design direction. The package format and cross-system portability are not finalized claims.

## The same context, where the work happens.
Browser collaboration and on-prem execution are deployment choices, not competing realities. Both work from the same engineering context.

**L3D Vue — Engineering Context Workspace**
A browser workspace where multidisciplinary teams and AI review 1D, 2D, and 3D against the same context. Evidence and CRS workflows stay connected to the objects and revisions under review.

**nanochat — A local AI agent on existing CAD workstations**
Combining SLM with Engineering Context, it understands drawings, 3D models, requirements, and change history on internal networks and in the field, supporting design review.

nanochat is in active development, with an Alpha release planned for December 2026.

## One context layer. Purpose-built products.
Each product owns a distinct part of the engineering context. None of them replaces the source of truth.

[Explore the product family](/en/products/)

## Useful thinking for engineering teams.
Short, practical notes on context, change, spatial reasoning, and industrial AI. Leave with an engineering insight, not a product pitch.

[Read the insights](/en/insights/)

## Our products were shaped by real projects.
Not a logo wall. These are the engineering pressures that shaped how we build.

[Contact us](/en/contact/)