In agents, on devices, in silicon, at the edge.
We are building the architecture that travels with it.
Cloud first. Device next. Silicon after.
The work has already begun.
Inference cost is now the binding constraint on enterprise AI. Every enterprise running agents at scale wants the same quality, at a fraction of the cost.
That is the door we walk through first.
Not quantization. Not pruning. Not a smaller transformer. A fundamentally different construction.
| Benchmark | Standard 1.5B | ANT 272M |
|---|---|---|
| LAMBADA (accuracy ↑) | 39.0% | 34.5% |
| HellaSwag (accuracy ↑) | 35.0% | 32.0% |
| WikiText PPL (lower ↓) | 26.6 | 32.8 |
5.6× fewer parameters. Same training compute. The gap closes as the models get larger. This raise extends the curve to 70B.
Each stage funds the next. Each stage de-risks the next. One vision, built in four stages.
EverydaySeries is our agent platform. Three paying customers today, running on frontier LLMs.
Beginning this raise: ANT models light up inside the platform. The same agent task, routed to our model instead of GPT or Claude — at a fraction of the inference cost.
The customer pays less. We capture the margin. Every query routed to ANT is revenue, evidence, and a data point for the next stage.
EverydaySeries adds a fine-tuning layer. The customer describes the task, uploads their SFT data, picks a device target — and gets back a purpose-built ANT model packaged with the device runtime.
Not a general model made smaller. A model built for one job, at the smallest size that job requires, deployed where the job happens. No cloud. No subscription. No data leaving the device.
The EverydaySeries customer who fine-tuned an ANT model for their cloud agent already has the data. Putting the same model on their device is the natural next sale to the same customer.
Already proven: a language model runs on Apple Watch — dual-core CPU, no Neural Engine, no internet. The architecture works at the smallest commercial silicon that exists.
We have been building this for five years. The market is now ready to buy it.
The cloud incumbents cannot pivot — they are committed to the wrong axis.
We are the company built around the right one.
Everything you have seen so far was built on this. The next stage is what this raise unlocks.
Raise closes the team. Commitments in place.
Either way, stage one is a real business. The later stages are upside, not survival.
We are not asking you to bet on the whole vision.
We are asking you to fund the first stage —
which is already underway.
We are not asking you to fund a frontier AI lab.
We are not asking you to fund another agent platform.
We are asking you to fund the architecture that takes intelligence everywhere — starting with the place enterprises are already paying.
Cloud now. Device next. Silicon after. License alongside.
Your question will be emailed to Gaurav. He'll respond directly.
Private to you. Not sent anywhere. Lost on refresh.