Build a custom model with your coding agent.

Agent-native TypeScript framework for open-weight models

Build a custom model with your coding agent.

Tell Claude Code or Codex what you want the model to do. Your agent prepares the data and TypeScript training workflow in your repo. You review the code; Arkor runs the training on managed GPUs and deploys the result behind an API your app can call.

Build your first modelTry Gemma 4 instantly

No signup required — without an account, your endpoint is created anonymously and expires automatically. By creating an endpoint you agree to the Terms of Service and Privacy Policy.

Built for TypeScript and Next.js teams. No separate Python training stack or GPU infrastructure.

$pnpm create arkor@alpha

Describe it. Review the code. Run the model.#

01

Tell your coding agent what model to build

Tell Claude Code or Codex the behavior you want, not the training configuration.

You, to your coding agent

“Build a model that rewrites rough drafts as tweets in my voice. Use Arkor.”

Your agent then

• Find or prepare a dataset
• Convert the data into the required format
• Write the TypeScript trainer
• Add evaluation

02

Your coding agent builds the training project

It prepares the dataset, writes the TypeScript trainer, and adds evaluation inside your repo. Review every change before training.

• Dataset preparation
• TypeScript trainer
• Reviewable configuration

03

Studio opens. You click Run Training.

Your coding agent runs pnpm dev and launches Arkor Studio at localhost:4000. Review the detected trainer, click Run Training, and follow the training progress in Studio.

`pnpm dev`Arkor Studio → localhost:4000

You control the run. Arkor handles the compute.

Tell your coding agent what model to build.#

Copy a prompt into Claude Code or Codex. Your agent prepares the TypeScript project; you review it, then Arkor trains and deploys.

Tweet editorSupport triageStructured extraction

You→Claude Code / Codex

Copy

Use Arkor to fine-tune a model that rewrites rough drafts as tweets in my style: https://github.com/arkorlab/arkor Find or prepare a suitable dataset, create the TypeScript training workflow, and tell me when it is ready to review in Arkor Studio.

1. 1Agent builds the project 2. 2You review the TypeScript 3. 3Arkor trains and deploys

Your model is ready to use

Arkor deploys the fine-tuned model behind an OpenAI-compatible API. Call it from your Next.js or Node app by changing the base URL.

Example output

Endpoint ready

Modelexample-fine-tuned-model

Base URLhttps://example-model.arkor.app/v1

API keyark\live\_example\••••••••

Example values shown. Your real endpoint and key are issued after deployment.

const client = new OpenAI({
  baseURL: process.env.ARKOR_BASE_URL,
  apiKey: process.env.ARKOR_API_KEY,
});

Fine-tuned model · Reviewable training code · Hosted API

Want to try an open model first?#

Get a temporary Gemma 4 API in one click. No signup, installation, or GPU setup required. Without an account, the endpoint expires after 7 days; sign up to keep yours.

Try Gemma 4 instantly

No signup required — without an account, your endpoint is created anonymously and expires automatically. By creating an endpoint you agree to the Terms of Service and Privacy Policy.

Base model only; no fine-tuning is applied.

Describe the model you want.#

Your coding agent builds the TypeScript project. You review it. Arkor trains and deploys it.

Build your first modelView on GitHub

Ask me anything.#

How to use open-weight models in your product, how to start fine-tuning, or how to ship personalized AI features for your users. I'm happy to help!

Book a quick chatDM me on X