Design Explorer

Explore a design space, not one guess.

Run parametric design optimisation on your Mac. Choose editable CAD dimensions and bounds, set objectives and engineering constraints, then let Lattiform rebuild, mesh, solve and compare candidates locally using the Metal-backed FEA workflow.

Design Explorer: in Lattiform now

Agent design tools: beta

In Lattiform now

The in-app Design Explorer is in Lattiform from the 19 October 2026 release: choose View > Workspace > Design Explorer. AI agents can run the same kind of exploration through the design tools in the MCP helper, as a beta. The windows below are real, rendered from development source shortly before that release.

Start 7-day trial

What ships today

An AI agent set up and ran this study over MCP, live in the open Design Explorer window: 40 wall-bracket designs, each rebuilt and solved by FEA, then plotted and compared. Real time, unedited, no sound. Recorded 15 October 2026 from Lattiform’s development source, before the 19 October 2026 release that includes it.

How a study works.

Step 1

Bound the parameters

Any editable dimension, fillet radius, hole diameter, shell thickness, pattern count or feature suppression can be a variable: continuous, integer, a set or a category, each with a range. Values driven by an expression are refused; permit the parameter they read instead.

Step 2

State objectives and constraints

Objectives with targets, weights and priorities; hard and soft constraints. Every limit in the engineering brief becomes a hard constraint. Goals typed in plain words become a draft you approve, and a unit is never guessed.

Step 3

Run the candidates

Each design is rebuilt in CAD, meshed and solved on the packaged Metal route. Preflight lists what blocks a run before any solve. Runs pause, cancel and resume exactly.

Step 4

Compare and decide

Scatter, parallel coordinates, correlation, uncertainty and history views. Compare up to four designs side by side, with their refusals, manufacturing verdicts and provenance.

Design Explorer Setup for the mounting-bracket demonstration, showing bounded radius and extrusion-distance parameters. — Actual app capture

Full app window. Enlarge opens the original window with zoom and pan.

01 / 04Setup

01 · Setup. Set bounded parameters for an illustrative study with a fixed underside and a 500 N lateral force. The study’s limits are 20 MPa stress and 0.02 mm displacement.

Loading capture: Setup…

Development preview from before the 19 October 2026 release. Step through captured states; playback does not represent solve time. Reduce Motion is on; steps change without movement.

When a candidate fails, it says why.

Refusals name the stage

Every design passes resolve, CAD edit, rebuild, remap, preflight, mesh, Metal solve, recovery, manufacturing and cost checks. A refusal names the stage and the reason.

Unknown never passes

A value that was not trusted or not measured is marked Unknown and never counts as meeting a constraint. Surrogate predictions only decide where to look; they are never shown as results.

No solver, no verdict

If this Mac cannot solve at all, the run stops with the reason, records nothing about the design, and resumes once the solver is available.

Design Explorer Report in a development build of Lattiform showing the mounting-bracket mesh-refinement sequence, retained mesh-sensitive warning and provisional ranking. — Actual app capture

Editable output, where supported.

Back into CAD

Reopen in CAD applies a chosen design to the open part as one undoable step, with every feature still editable. It is refused if the part has changed since the study was set up.

Exported and read back

Export a design as STEP, STL or OBJ; the file is read back and checked against the exact solid. Results export as CSV, plots, Markdown and HTML reports with a checksummed manifest.

Reproducible

A study keeps the CAD state it was written against, so editing the part never strands it. A paused or interrupted run replays its record and refuses to continue if anything would differ.

Design Explorer results in a development build of Lattiform plotting mounting-bracket mass against peak von Mises stress, with nine evaluations from seven unique geometries and two finalist mesh refinements. — Actual app capture

Available now

What you can use today.

The MCP helper inside Lattiform gives an AI agent design tools: an engineering brief, exploration, candidate proposals and previews, evaluation, convergence, sensitivity and sweeps, comparison, and adoption into the part with rollback.

How autonomous is it?

There is no fully autonomous designer in the app yet. Autonomy today means an AI agent driving these tools over MCP, or the Explorer’s search running within the bounds you set. Every candidate is still checked against the brief’s limits, and nothing is adopted into your part without an undoable step.

Connect an agent

Start 7-day trial

Status.

Design Explorer

Shipped

In Lattiform now

Vary a part's dimensions, fillet radii, hole sizes, shell thicknesses, pattern counts and feature suppression within bounds; solve every candidate on the Metal route; judge them against objectives and hard constraints; compare and keep the evidence.

Studies use FEA Lab's regular grid of cube elements and one catalogue material. Only a verified design can be exported from it. No published verification evidence.

Design tools for agents

Beta

In Lattiform now

An engineering brief, exploration, candidate proposals and previews, evaluation, convergence, sensitivity, sweeps, comparison, adoption and rollback, and Design Explorer studies, through the MCP helper.

Every candidate is checked against the brief's limits; a refused or failed candidate is reported with its reason. No published evidence.

Autonomous design loop

In development

Not in a released version

No feature by this name exists in the app. Today autonomy means an AI agent driving the design tools over MCP, or the Explorer's search running within the bounds you set.