Private beta

Define. Fine-tune. Deploy.

Describe your task — we'll generate the data, train the model, and export it for any platform. You bring the use case, we handle the ML.

Join the waitlist →
  • No ML expertise
  • No GPU setup
  • Your data stays yours
How it works

From a use case to a model that's yours.

Forgelet runs the whole pipeline. You pick the task and define it — we generate the data, fine-tune a small model, and export it in your format. No notebooks, no MLOps, no lock-in.

Forge line forging
01 / SELECT

Pick your use case

Function calling, classification, or data extraction.

02 / DEFINE

Define your task

Define tools, categories, or fields — a short description works, or a JSON spec for tighter control.

03 / DATA

Add data or generate it

Bring your own dataset, or let Forgelet synthesize training data from scratch. Or combine both.

04 / FINE-TUNE

Fine-tune in the cloud

A small base model is trained on managed GPUs. No notebooks, no MLOps.

05 / YOURS

Download & deploy

Export to GGUF, LiteRT, or CoreML. Yours to run anywhere, offline, forever.

Before / after

The fine-tuning journey — before and after.

Without Forgelet
  • Research frameworks & tooling
  • Provision GPUs & environment
  • Collect & clean your data
  • Format datasets (JSONL, etc.)
  • Configure training runs
  • Debug, tune, and iterate
  • Convert to deployment format
  • Fix export & runtime errors
  • Validate on target device
Typical timelineDays–weeks
With Forgelet
  • Define your task
  • Add data or let Forgelet generate it
  • Download a model file ready to deploy
Typical timeline~ 20 min
Use cases

What you can build.

Function Calling

Your app understands natural language and calls the right function. Locally, offline, on any device.

input"Turn off the bedroom light"
↓
output
{
  "name": "toggleLight",
  "arguments": {
    "room": "bedroom",
    "state": false
  }
}
Classification & Routing

Sort messages, tickets, or reviews into categories. Route support requests to the right team automatically.

input"I want a refund for my order"
↓
output
{
  "category": "returns"
}
Data Extraction

Pull structured data from messy text. Names, numbers, amounts, dates — extracted in one pass.

input"Call me at 555-0123 tomorrow at 3pm"
↓
output
{
  "phone": "5550123",
  "time": "15:00"
}
Role & Persona

Teach a model a specific tone, style, or character. Brand voice, support agent, domain expert — consistent every time.

input"What's your return policy?"
↓
outputAnswers as your support agent, using your rules and tone.
Own the model

Why rent a model when you can own one?

A well-tuned small model can match — or even outperform — models many times its size on your specific task. And unlike cloud APIs, it doesn't charge per token or send your data to someone else's servers. It's yours, it's free to run, and it fits where your app actually lives.

01 / yours

It's yours

Download the model. Run it on your phone, your server, or your Raspberry Pi. No vendor lock-in.

02 / free to run

It's free to run

No per-request pricing, no token billing. Train once, run forever.

03 / private

It's private

User data never leaves the device. No third-party calls to explain in your privacy policy.

Early access

Get your hands on Forgelet first.

Tell us what you'd build. We're onboarding in waves.

We'll only email you about access — nothing else.

You're on the list.

We'll reach out as access opens in waves. Thanks for the interest.

Questions

Good things to know.

Do I need ML experience?
No. Describe the behavior you want — or pass a JSON spec — and Forgelet handles dataset generation, fine-tuning, and export for you.
What hardware can run the model?
Export to GGUF (desktop and server), LiteRT (Android), or CoreML (iOS). The model file runs fully offline on the target device — no network, no per-request cost.
Is my data private?
Your data trains your model and isn't pooled into shared models or sold. You can run the result entirely inside your own perimeter.
What model sizes do you support?
Small on-device bases — roughly 270M to 3B parameters — sized to run on a phone. We pick one that fits your target device.
How does pricing work?
You pay once to forge a model in the cloud; running it on your own hardware has no per-request cost. Details firm up closer to launch.
Forging soon

Forge your first model.

Drop your email. Early access opens in waves.

One launch email, no spam.