Specialized models.
Factory scale.

The S Series is our first production reference. LIVIA is a family of specialized legal models running on Grid inside Forlex. In August 2026, the company reported 240,000 monthly active users for Forlex. The reference page identifies the source, the workflow and the controls specific to that legal product.

The differentiator is not only legal training. It is verifiability. OpenDetector screens every input for prompt injection before it reaches the model. OpenJuris checks every output for hallucinations and unsupported citations before it reaches the lawyer. OpenContracts verifies that generated text stays faithful to the source document. Three layers, each auditable. They are how Forlex uses Grid. They are not Grid itself. Grid is the compute, the routing, and the cost model underneath.

Forlex is one vertical, not the company. We built the legal product because that is the market we know, and because it forced the factory to be honest. The next products, ours or a partner's, reuse the same line: high throughput, low latency, predictable cost, and an audit trail a stakeholder can read.

The company was founded in 2023 in Goiânia, Brazil. Specialized models should run at factory scale, with named routing. The S Series shipped first, for law. Grid is the line those models, and others, run on. Forward Intelligence is the company behind both. The team is small, senior, and cost-effective. We run close to the metal because that is how you keep latency down and unit economics sane.

We are pursuing ISO/IEC 42001 (AI management systems) and ISO/IEC 27001 (information security). We have not claimed the certificates. When they land, they land in writing. We do not sell an open model marketplace. Every model in the catalog earns its place. We do not publish prices we cannot stand behind. We quote by volume when a number is not settled yet. We do not invent a hallucination rate we cannot source.

In August 2026 we expanded our AWS commitment to US$77 million. Hundreds of NVIDIA B200 GPUs are online. More than 8,000 additional GPUs are planned by year-end. Partners include NVIDIA, AWS, Cloudflare, Linode Rise, and AB2L. We were selected for the AWS Generative AI Accelerator 2025: one of three Brazilian companies in a global cohort of forty. That is the plant we operate today. Grid is the production line.

Grid continues to run on AWS. That is the current plant. We are also developing two expansions: one site in Brazil and one in the United States. The Brazilian site is a 20 MW expansion expected between 2026 Q4 and 2027 Q1. The US site is expected between 2027 Q3 and Q4 and supports 10 MW. In Brazil we expect to deliver 500 MW of power capacity in phases from 2027 to 2029. Those sites are in the pipeline. They are not the factory Grid runs on today. The sites section below labels each status so a pipeline is not read as a live factory.

Grid inference is live, invite-only, on the AWS plant. Fine-tuning is by invitation. Tune it where you serve it. Custom training, the Training API, and managed jobs are the three paths. If you are building a product that has to survive thousands of concurrent requests, and you need the bill and the routing to make sense to an executive, start there. If you want us to design and run a custom project on the factory with you, talk to us. That is a workload on Grid, not a training job and not a site. If you are planning capacity in Brazil or dedicated halls in the United States, talk to us about the sites.