NewThink Lovable, but for training AI models

Post an idea. Get your own AI model.

Chat with Kind Postbox, a friendly expert. It asks a few questions, builds and checks the training data, trains a small model and hands you an API.

Plans from $29 a month. Cancel any time.

Plain English, start to finishYou check its answers before it trainsAn API in one click, or download the model

How it works

You chat. Kind Postbox does the heavy lifting.

1

Post your idea

Describe the task in a sentence, like "sort support emails by urgency". Add example rows if you have them.

2

Answer a few questions

It asks what the labels should be and what decides them, one question at a time.

3

Check 20 answers

Grade 20 of its test answers without seeing its picks. Your corrections, and your reasons, fix the whole test set.

4

Get your model

It gathers and writes training data, trains a small model, scores it honestly and gives you an API.

Kind Postbox
Question 1 of 2

What urgency labels should the model use?

13 · Urgent, Normal, Low23 · Urgent, High, Normal32 · Urgent, Not urgent

One question at a time. Never more than three before it starts.

Check 20 answers7 of 20

"I can't log into my dashboard. It says my account is temporarily suspended for security reasons. I need access…"

1 · Urgent2 · Normal3 · Low

You won't see its answer until you've picked yours. Twenty rows, once, and you're done.

Honest results

Honest numbers, before you ship.

Every version is scored on the test set you checked, next to a model that wasn't trained, and the misses are spelled out in plain words. You decide when it's good enough.

Real numbers from a live Kind Postbox model, measured on the 107 test emails its owner approved.

Support email urgency · v5Live
Urgent caught96%Target 92% met
All correct72%Target 85% not yet
Untrained model30%all correct
Where it still misses
Low emails read as Normal
19 of 40
Normal read as Low
7 of 40
Urgent missed
1 of 27
Cost so far $0.15 · 5 versions

What you can build

One sentence. Nine kinds of model.

Kind Postbox reads how you describe the task and picks the right recipe. You never choose a model type.

One label"Sort emails into Urgent, Normal or Low"
Several tags"Tag every emotion in a message"
Category, then subcategory"Credit card › Store card"
A level on a scale"Rate how respectful a comment is, 0 to 4"
A number"Score each comment from 0 to 100 for toxicity"
Other languages"Sort requests in English, French and Swahili"
Long documentsBeta"Which risky clauses does each contract have?"
Two texts togetherBeta"Grade an answer against its question, 0 to 4"
Pull pieces out of textBeta"Find product and company names in posts"

Ship it

An API in one click. Or take the files.

Publish puts your model live behind an API key. On paid plans you can download it, with a script that gives the same answers, and run it on your own servers for free.

curl -X POST https://api.kindpostbox.com/v1/models/<id>/predict \
  -H "Authorization: Bearer tk_••••••••••••" \
  -d '{"inputs": [{"kind": "text", "text":
       "My payment failed and my account is locked"}]}'

{"answers": [{"kind": "category", "label": "Urgent"}]}
Download the modelModel files plus a script that gives the same answers as the API

Pricing

Simple plans. Hard limits. No surprise bills.

Every limit is checked before anything runs, so a build never costs more than your plan.

Starter

$29a month

  • 5 models a month
  • 2.5 hours of training
  • 20,000 predictions
  • 3 live models
  • 3 model downloads
  • 1 seat
Choose Starter

Pro

$99a month

  • 25 models a month
  • 10 hours of training
  • 250,000 predictions
  • 15 live models
  • 20 model downloads
  • Bigger models and long documents
  • 3 seats
Choose Pro

Team

$299a month

  • 100 models a month
  • 40 hours of training
  • 2,000,000 predictions
  • 50 live models
  • 100 model downloads
  • Bring your own base model
  • 10 seats
Choose Team

Pay yearly and get 12 months for the price of 10.

Questions

Do I need to know machine learning?

No. You describe the task in plain English and answer a few questions. Kind Postbox picks the recipe, the model and the training settings.

How is this like Lovable?

The same idea, a different result. Lovable turns a prompt into an app; Kind Postbox turns a prompt into a trained AI model with an API. Kind Postbox isn't made by or connected to Lovable.

What can it build today?

Text models: one label, several tags, category then subcategory, a level on a scale, a number, and other languages. Long documents, comparing two texts and pulling pieces out of text are in beta.

Where does my data go?

While it builds, your idea and examples are sent to DeepSeek, the language model that writes and checks the training data. Training runs on cloud GPUs. Your models and files stay in your account.

Who owns the model?

You do. On paid plans you can download it, with a script that gives the same answers as the API, and run it anywhere.

What if the model isn't good enough?

You see an honest score on your own checked test set before you publish, and where it still misses. Keep improving with another version, or stop. Every build stays inside your plan's limits.

Can I cancel?

Yes, any time from your billing page.

Your next model is one sentence away.

Post your idea