Classify anything against your own labels.
Point it at your schema and it starts making decisions immediately. No labeled dataset, no training step, about 30ms per call, running on hardware you control.
I was charged twice for my annual plan and I need one of the charges reversed before Friday.
- Billing
- Bug report
- Cancellation
- Feature request
Duplicate charge on a subscription, routed straight to the billing queue.
Examples are illustrative. Latency is the published P50 on a single GH200.
Start instantly. Turn the dial when you need more accuracy.
Your schema, day one
Define your labels and start classifying. No labeled dataset, no training run, no ML team on the critical path.
A small labeled set, a big jump
Add examples and fine-tune in your own environment with P95. Weights never leave your hardware.
Published accuracy figures reflect the tuned (P95-trained) state, not the no-training baseline.
Fast enough for the hot path. Consistent enough to trust it there.
0.6B model on text-classification benchmarks. All runs self-hosted on one GH200, in-environment.
Every team has text that needs a decision.
Tickets and inbound email
Send every message to the right queue the moment it arrives.
Compliance and policy
Catch risky language in communications before it becomes an incident.
Documents
Sort contracts, filings and correspondence by type and priority.
Intent
Separate buying signals from support noise across every channel.
Content
Apply your own policy taxonomy, not a vendor default.
Events and transactions
Label high-volume records at pipeline speed without a model call to an outside API.
Runs where your data already lives.
Your text never has to leave your environment to get classified.
Your cloud
Deploy inside your own VPC on GPUs you already pay for.
Your data center
Fully air-gapped, on-prem. Built for regulated environments.
Our GPUs
No hardware yet? Run on Lupine, billed per second.
One platform. Use the layers you need.
The classifier runs on Numerata’s own stack: GPU orchestration, an agentic fine-tuning platform, and a custom inference runtime. Use all three or plug one into what you already run.
Join the beta. See decisions on your own data.
For teams deploying in-house
We’re onboarding a small group of teams to run the classifier on their own labels. Tell us where to reach you and our head of product will follow up.
Thanks, you’re on the list. Look out for an email from our head of product.
For developers
Spin up GPUs on Lupine and start building today.
Beta spots are limited. We reply within a day, from our product team.