Ora Frontier is now Sparset

If you knew us as Ora Frontier, you are in the right place. Our name is now Sparset, and our focus is sharper than it was. Sparset builds the software that puts AI models into production and keeps them running efficiently, on infrastructure our customers control.

This first post covers where we started, why we changed, and what that means now.

What Ora Frontier set out to do

Ora Frontier built private AI for your own hardware. The idea was simple: bring AI closer to people and their work, instead of sending every file and prompt to someone else’s cloud.

We split that idea into three products. Core was a personal AI workspace on your own computer. Stack offered specialized models and optimized inference for products. Edge deployed and operated AI across private servers and controlled environments.

Why we changed

Three products meant three audiences: individuals, product teams, and enterprises. Explaining all of them at once made it hard to say plainly what we do.

A customer made the answer clearer. Roomates came to us while we were still Ora. They didn’t need a desktop app. They needed their custom avatar model running quickly and affordably in production. Their monthly serving costs fell from $2,000 to $400 at 50,000 generations, at about three seconds an image. Read the Roomates story.

That kind of work does not end at launch. A production model has to stay fast, reliable, and affordable every day after. Done in-house, that means a platform layer your engineers have to build and maintain. Sparset takes that work on, and your team keeps the final say.

Why “Sparset”

It’s a play on “sparse set.” In computer science, a sparse set keeps track of only the items that are actually there. Sparsity in AI models follows the same instinct: store and compute only what matters, and the same hardware does more. That instinct runs through our work.

What changed

• Who it’s for. Teams and enterprises running AI models on their own servers and in their own data centers.
• What it is. An inference operations platform. Agents plan, optimize, deploy, and operate your model as one continuous job.
• Core is retired. No more personal AI workspace.
• Stack and Edge have new jobs. Stack operates production inference on GPUs you own or rent. Edge does the same under stricter governance, control, and on-premises requirements.
• Forge is new. Forge builds custom and fine-tuned models around your tasks. Stack or Edge can operate them.

What stayed the same

• Your infrastructure. AI running where you control it is still where we start.
• Your call. You decide what runs on its own and what waits for approval. Every change that needs you comes with its reason and its way back.
• Your payloads. Production requests and responses stay in your environment.
• The research. Making inference more efficient is still one of our two lines of work.

Where to go from here

If you run models on your own infrastructure and keeping them running eats more of your team’s week than it should, we’d like to hear about it.