Platform

One model layer, one data store, five tools on top

Forgite is a platform underneath and products on top. Every product reads the same potentials and writes to the same provenance graph, so a measurement in one place sharpens a prediction everywhere.

Abstract

Layers
6
Products
5
Deploy
Cloud, VPC, on-prem
Status
Pre-launch

01 · Architecture

Six layers, from data to the lab bench

The architecture of the first release. Generative inverse design and process digital twins are later additions.
LayerWhat it doesBuilt onProduct
DataCurated DFT and MD datasets, your experimental results and literature, kept per tenantObject store, RAPIDSAll
Foundation potentialsTrain and fine-tune equivariant interatomic models with calibrated uncertaintycuEquivariance, DGX/HGXEquiva
Simulation and screeningRelaxation, hull distance, MD and property calculators at batch scaleWarp, ALCHEMI, TritonDynara
Active learningChoose the next simulations or experiments by expected informationCUDA, RAPIDSInventa
Discovery consoleRankings, property views, provenance, research copilotNeMo, NIMMentra
Lab integrationJobs to partner or in-house automated labs; results back into the data layerJetson, HoloscanSynthex

02 · Flywheel

Every campaign leaves the models better than it found them

1MentraResearch copilot

Reads the literature and your notebooks, frames the question and drafts the campaign.

2InventaDesign + active learning

Turns the target into candidates and picks the next batch by expected information.

3EquivaInteratomic potentials

Predicts energies, forces and stresses at near-DFT accuracy, with an uncertainty for each.

4DynaraGPU simulation

Relaxes, scores against the hull and runs GPU molecular dynamics on the survivors.

5SynthexSelf-driving lab

Sends the short list to partner or in-house labs and brings measurements back.

Return pathSynthex measurements → fine-tune Equiva → better predictions in the next round.

03 · Deployment

Where it runs

Option 1

Forgite cloud

Managed GPUs, isolated per tenant. The fastest way to start a pilot.

Option 2

Your VPC

The platform deployed in your cloud account on your GPU quota. Data never leaves your account.

Option 3

On-premises

For IP-sensitive programs: NVIDIA AI Enterprise on your DGX or HGX systems, updated by signed bundles.

04 · Control plane

What your team controls

Search spacesWhich prototypes, elements and substitutions a campaign may explore.
Filters and targetsHull threshold, property limits, excluded elements, cost or supply rules.
Compute budgetsGPU-hours per campaign, with a hard stop and an alert at 80 percent.
ApprovalsWho signs off a batch before it goes to synthesis.
Model versionsPin a potential version for a campaign so results stay reproducible.
Data sharingOpt in or out of contributing results to the shared foundation model.

05 · Security

Built for unpublished chemistry

  • Tenant isolation for data, fine-tuned models and logs.
  • Encryption in transit and at rest; customer-managed keys in VPC and on-premises deployments.
  • Single sign-on and role-based access; every action recorded in an audit log.
  • Independent security audit (planned before general availability).

Read the validation and security notes

06 · Questions

Platform questions

Can we bring our own potential?

Yes. Dynara and Inventa accept any potential served behind the same interface, so you can compare it with Equiva on your own benchmarks.

What happens if a model version changes mid-campaign?

It does not. Campaigns pin the potential version they start with; a new version starts a new campaign that you can compare side by side.

Next step

See the platform on your chemistry

A pilot runs on one program, with your data in your tenant, and ends with a written report of what worked and what did not.