Wity for Enterprise
Your Model, Now A Decision Model.
Bring the model you already run and we turn it into a System 1.5 decision model, or run Wity-1 itself. Either way you get typed answers with a probability on every option, calibrated on your data and running inside your network. Talk to us about your model, your data requirements or higher limits.
your model
open weights
wity ▸
decision
approve this refund?
- approve0.00
- review0.00
- reject0.00
→ { choice: "approve", p: 0.94 }
Two Ways To Run It
Wity-1, Or Your Own Model.
Run Wity-1, the model behind every benchmark on this page, or bring a model you've already chosen and approved, and we turn it into the decision model. Either one runs inside your network or as a dedicated instance in our cloud, behind the same API and SDKs.
Wity-1
Our System 1.5 model, as benchmarked below.
- ✓On Wity Cloud now, through the public API
- ✓As a dedicated instance or on your hardware
- ✓Calibrated on your data for your deployment
Your Model
A model you already run, made to decide. We're not tied to any one model.
- ✓Open weights, a fine-tune or a model built in-house
- ✓The weights never leave your network
- ✓Measured on your data before you switch
Bringing your own model
- 01
Bring your model
Tell us what you run and what it needs to decide. We scope it with you.
- 02
We make it decide
Typed answers with a probability on every option, behind the same API and SDKs as Wity Cloud.
- 03
We calibrate it
On your labelled examples, so its probabilities hold up on your traffic.
Proof · Wity-1
Replace Your Expensive LLMs.
Most decisions inside software don't need a chat model writing paragraphs. Every number in this section is Wity-1's. A model you bring gets its own numbers, measured on your data before you switch.
Independent · Benchmark Heaven
Top Capability Score Among Jev-Class APIs
Ahead of Sage 1.3.0 at 78.6 and Jev 1.13.0 at 77.1, on 1,500 decisions, 1,200 of them sealed.
Jev-class: at most 2× Jev's cost and 2.6× its median latency, 12 of 23 APIs. 4th of 23 on the official composite.
JevBench Capability Score ↗First On Image JevBench
First of 50 systems on the official composite, with 89.9% accuracy on sealed image decisions.
Second place, Imajev-4B, scores 76.39. 684 image decisions, 456 of them sealed.
Image JevBench ↗Our benchmark · 8 fast, low-cost models · 200 real-world cases
Faster In Model Time
109 ms typical call
Next fastest: 943 ms
Cheaper Per Run
$3.29 per 1,000 runs
Next cheapest: $17.95
Averages across all eight models.See the benchmarks
On Wity Cloud
You pay per token
$0.042 per 1M tokens, and failed calls are free. The cost figures above are Wity Cloud prices.
On your GPUs
You pay in GPU time
There's no per-token bill. Your cost is the hardware, and a decision in one pass takes a fraction of the GPU time of a model writing out its answer, so the same GPUs serve far more decisions.
Calibration
Calibrated To Your Decisions.
Out of the box, a decision model's confidence shows how sure it is, not how often it's right. In an enterprise deployment we calibrate it on your own examples, so when it says 0.9 it is right about nine times in ten on your data, and the thresholds you act on hold up in production.
Your labels
Your classes and your past outcomes, not a generic benchmark.
Your thresholds
Where to act, where to escalate and where a person should look.
Your cost of a mistake
A missed fraud flag and a false alarm don't cost the same. The calibration knows.
how often right →
Illustrative. Dashed squares mark perfect calibration.
Latency
No Network In The Way.
Wity runs next to the software that asks the question. There's no round trip to a provider and no shared queue, so the time you wait is the time the model takes.
Hosted LLM API
answered ✓
connect · upload · queue · model · download
Wity in your network
answered ✓
model
Illustrative. On a hosted API, each segment varies with region, payload and load.
Your Data
Your Data Stays Yours.
content stored: none
Zero Data Retention
On every endpoint, by default. Your requests and answers are processed in memory and never logged, written to disk or backed up. We keep only call metadata, such as time, status and token counts, to bill you.
Data Retention Policytraining: off
No Training On Your Inputs
We don't use your prompts, images or answers to train or fine-tune models, ours or anyone else's.
Privacy Policyegress: none
Inside Your Network
Run on your premises or in your own cloud. Requests, answers and your model's weights never leave your environment.
Talk to usFAQ
Enterprise FAQ
Yes. Bring the model you run, whether open weights, a fine-tune or one trained in-house, and we turn it into a decision model that answers with typed options and a probability on each. We're not tied to any one model. We measure it on your data before you switch, so you see its own numbers, not Wity-1's. Tell us what you run and we'll scope it with you.
Calibration is part of an enterprise deployment; confidence on the public API is uncalibrated. We use your labelled examples, past decisions and their outcomes, to tune the decision model so its probabilities match how often it is right on your data. Then we set your thresholds for acting, escalating and asking a person around what a mistake costs you.
No. We don't use your inputs or outputs to train or fine-tune models.
Yes, on every endpoint, by default, with nothing to set up. Request and answer content is processed in memory and is never logged, written to disk, backed up or used for training. The model server may hold recent requests briefly in an in-memory cache to answer repeats faster; later requests overwrite it. We keep call metadata (time, status, token counts and cost) so we can bill you, never the content. On-prem, nothing reaches us at all.
On Wity Cloud, requests pass through our API gateway to the GPU server that runs the model, in memory only. A data processing agreement (DPA) is available on request. On-prem or in your own cloud, everything stays inside your environment.
On Wity Cloud it's $0.042 per 1M tokens, pay as you go, and failed calls are free, with committed-volume rates for large workloads. On your own GPUs there's no per-token charge: the deployment is priced as a whole, and the compute is yours. Talk to us for either.
Contact
Let’s Talk Business.
Tell us what model you run, what you need to decide and where your data has to stay. We'll come back with a plan.