Build your model

Custom small language models

Give your work
its own model.

We turn your domain knowledge, documents, structured text, or source code into a focused AI model. You download it, run it locally, and own every file.

Build your model See how it works

Private beta · Workspace subscription + transparent training credit

YOUR MODEL READY
claims-expert.gguf 3B · 2.14 GB
TRAINED FOR

Classifying claims urgency and returning the approved JSON schema.

  • 01model.ggufYOUR MODEL
  • 02evaluation.pdfTEST RESULTS
  • 03quickstart.mdRUN LOCALLY
Works with Ollama · llama.cpp

Runs locally. Your data stays yours.

Clear costs. Approve every training run.

Open format. No vendor lock-in.

How it works

From raw material
to a model in three steps.

01

Define the behavior

Describe the exact input and output: classify, extract, summarize, transform, generate, answer, or follow your own text instruction.

02

Share the source

Upload labeled JSONL examples or supported documents, tables, structured text, and source code. We normalize them into training data.

03

Download your model

We train, test, and package it. You receive a GGUF model that runs with Ollama, llama.cpp, and compatible tools.

What makes it different

We don’t host your model.
We hand it to you.

Your subscription organizes the workspace; it does not lock the finished model. Put the GGUF in your product, run it on a laptop, or keep it inside your network. There is no inference API key and no query meter running in the background.

FORMAT
GGUF
MODEL SIZE
1.5B / 3B / 7B
RUNTIME
OLLAMA / LLAMA.CPP
OWNERSHIP
100% YOURS

Product model

Train often.
Keep every version.

A subscription keeps your workspace active. Prepaid credit covers the variable cost of each approved training run.

Workspace subscription

Workspace

Organize multiple projects, models, versions, runs, and artifacts.

  • Project-based document library
  • Multiple model identities and versions
  • Training history and evaluations
  • GGUF artifact library
Create workspace

Training credit TRANSPARENT USAGE

Credit

Fund knowledge generation and GPU work without an open-ended bill.

  • Estimate shown before every run
  • Credit reserved only after approval
  • Actual cost settled in a visible ledger
  • Purchased credit does not expire
Open a workspace

Questions

Before you train.

Does my model need the cloud?

No. The finished model runs on your own hardware. Once delivered, it has no dependency on WeTrain or a hosted API.

What can I train it to do?

Any focused text-to-text behavior that can be demonstrated with inputs and expected outputs: labels, JSON extraction, summaries, transformations, generation, Q&A, style, support, or code.

What kind of data can I use?

Use labeled JSONL messages or input/output pairs, documentation, Markdown, CSV or TSV tables, HTML, XML, JSON, text files, and source code. Image, audio, PDF, and ZIP processing are not beta-ready yet.

How long does it take?

Each run will show an estimate before you approve it. We have not published a delivery SLA before completing real GPU benchmarks.

What if the result is not good enough?

Artifacts remain locked until they pass a held-out quality gate. The retry, review, and refund policy is still being finalized before paid beta.

Private beta

One model.
One job. Done well.

WeTrain is preparing its first customer projects. The order flow will open to early-access teams first.

Create your workspace