nebius/token-factory-cookbook

What it solves

Nebius Token Factory Cookbook saves you from starting from scratch when building applications with open models on the Nebius Token Factory platform. Instead of hunting through separate docs and examples, you get a single set of ready-to-run recipes that cover common tasks like making API calls, building agents, adding tools, doing retrieval-augmented generation (RAG), evaluating models, and even adapting models after training.

How it works

You clone the repository, copy a sample environment file, add your Token Factory API key, and choose a model from the live catalog. Each recipe lives in its own folder with its own dependencies and run instructions. Many recipes use popular frameworks or SDKs (such as OpenAI-compatible client, LangChain, LangGraph, Mastra, or Next.js) and show a concrete, working example. A quick start guide and an environment setup file get you moving, and a notebook version is available for the native API. You can browse by what you want to do — compare models, build an agent team, add tool calling, search over your data, or run a full application — and each recipe’s README explains required keys, optional services, validation, and deployment assumptions.

Who it’s for

Developers who want practical, copy-pasteable examples for using open models on Nebius Token Factory — whether you’re new to the platform or already familiar with it. It’s also useful for teams that need a starting point for a specific type of application (agents, RAG, coding agents, post-training, etc.) and for people who like to learn by running working code before adapting it.

Highlights

• Runnable examples with clear setup instructions for each recipe.
• Covers a wide range of use cases: API calls, agents, tool calling, RAG, full applications, integrations, coding agents, evaluation, post-training, distillation, and creative examples.
• Featured recipes include OpenCrewBot (a multi-bot chat app), Splitter Lite (a receipt-reading bill splitter), KubeTrace AI (Kubernetes incident investigation), a deep research agent, an internal support agent with retrieval, and a typed decision model.
• Includes a contribution guide and a recipe template so you can add your own examples.
• MIT licensed, with third-party assets retaining their own licenses.