Arkor

Arkor is a TypeScript-first platform for building, training, and serving custom open-weight models with a coding agent. Managed GPU training and OpenAI-compatible endpoint.

AI Agent Development

AI Developer Tools

Programming Agent

Visit Website

Overview

Arkor is a platform for building, training, and serving custom open-weight models through a TypeScript workflow. Instead of asking users to assemble a separate Python training stack and GPU infrastructure, Arkor lets a coding agent prepare the training project, after which Arkor runs the fine-tune on managed GPUs and deploys the result behind an API.

The product combines an open-source framework, a hosted platform, and a GitHub review bot. The framework handles TypeScript-based training runs and checkpoint deployment, the hosted product provides OpenAI-compatible endpoints and managed fine-tuning, and the code review app checks pull requests for stale comments and documentation drift.

Core capabilities

Coding-agent-driven setup

Describe the behavior you want in Claude Code or Codex, and Arkor turns that request into a TypeScript training workflow inside your repo for review.

TypeScript-first framework

Training runs live as TypeScript files, so you can version them, rerun them on new data, and roll back to earlier checkpoints.

Managed training on GPUs

Arkor runs training on managed GPUs and streams progress back while the run is executing, removing the need to provision your own GPU stack.

Checkpoint-based deployment

Each training run produces checkpoints that can be promoted to an OpenAI-compatible endpoint with its own URL and API keys.

Traffic-to-training loop

The hosted platform can turn requests and responses into runs that you can replay, inspect, and convert into training data.

GitHub review bot for drift detection

The code review product flags stale comments and documentation drift in pull requests without running repository code.

Practical use cases

Task-specific fine-tuning
Build a model that performs a specific transformation, such as rewriting drafts in a brand voice or converting support notes into structured output. Arkor's agent workflow prepares the dataset, trainer, and evaluation code for review.
Iterative improvement from live traffic
Use production requests and responses as training data, replay the runs, and turn them into a dataset for the next iteration. This fits teams that want the model to improve from real usage rather than from a one-off dataset.
App integration through API swap
Serve a fine-tuned model behind an OpenAI-compatible endpoint and connect it to a Next.js or Node app by changing the base URL. This is aimed at teams that already have model calls in production and want a custom model without changing their app architecture.
Pull request drift review
Install the GitHub App to catch comments and documentation that no longer match the code after a pull request. The bot posts review comments without blocking the check or running repository code.
Early experimentation with open models
Start with the temporary Gemma 4 endpoint to evaluate the workflow before committing to a fine-tune. The hosted open-model endpoint is intended for trying the platform without setup.

Pros and Cons

Pros

Lets a coding agent generate the training workflow so the user starts from a reviewable TypeScript project.
Runs training on managed GPUs, reducing the need for separate infrastructure setup.
Produces checkpoints that can be promoted, inspected, served, or rolled back.
Provides an OpenAI-compatible endpoint that can be swapped into apps that already call model APIs.
Includes a separate GitHub App for detecting stale comments and documentation drift in pull requests.

Cons

Pricing has not been published yet, so long-term commercial terms are unclear from the current pages.
Some capabilities are marked as coming soon, including serving without the SDKs, distillation, and confidential computing.
The source pages do not provide full documentation for integrations beyond TypeScript/Node and OpenAI-compatible API usage.

FAQ

How do you build and train a model with Arkor?

Arkor is built around a coding-agent workflow. You describe the model you want to Claude Code or Codex, review the generated TypeScript in your repository, and then Arkor runs training on managed GPUs before serving the result behind an API.

What languages or workflows does Arkor support?

The source pages show support for OpenAI-compatible endpoints and a TypeScript workflow for training and deployment. They do not document a dedicated Python SDK or non-TypeScript training path on the pages provided.

Do I need to manage GPU infrastructure myself?

The product pages say training runs execute on managed GPUs, while the framework keeps the training code in TypeScript files in your repo. You review the code before the run and can inspect checkpoints after training.

What does Arkor return after training?

Arkor's hosted platform deploys fine-tuned models behind an OpenAI-compatible API, and the framework page says you can promote a checkpoint to an endpoint with its own URL and API keys.

Is Arkor currently free?

The pricing page says Arkor is free during the alpha and that pricing has not been published yet. The hosted endpoint created without an account is described as temporary and expiring automatically.

Quick Facts

Category: AI developer tool
Platform: TypeScript framework and hosted model runtime
Primary workflow: Describe the model to a coding agent, review TypeScript, run training, deploy behind an API
Hosting: Managed GPUs and OpenAI-compatible endpoints
Pricing: Free during alpha; pricing not yet published
Source domain: arkor.ai

Arkor Alternatives

Termo

Termo is an AI agent platform with dedicated VMs for browsing, code execution, memory, and scheduled automation for research and build tasks.

Orca

Orca is an Agent Development Environment for shipping with coding agents, running multiple CLI agents in parallel across isolated worktrees, with desktop and mobile workflows.

Firebase Studio

Firebase Studio is a web-based workspace for full-stack app development with Gemini-assisted coding, app previews, cloud emulators, collaboration, and browser deployment.

AI Magicx

AI Magicx is a unified AI workspace for chat, image, video, voice, music, email and developer tasks, helping teams and creators manage multiple models in one place.

Paper

Paper is a design tool that connects canvas, code, and AI agents so teams can create, share, and ship work in one workflow. Includes desktop app and MCP access.

blop

blop is a QA agent that writes browser tests as code in your repo, runs them in CI, clusters repeated failures, and can open PRs to fix broken tests.