# kibbu

> kibbu serves your AI traffic on the computers your company already owns - cloud is the overflow, not the default. It runs inference on idle laptops, workstations and servers behind an OpenAI-compatible API, and sends what your fleet can't serve to OpenAI or Anthropic, priced and logged.

kibbu is built for the AI work nobody is waiting on: agents, evals, embeddings, nightly jobs and batch processing. It runs on macOS, Windows and Linux. Change one base URL, keep your OpenAI client.

## Start here

- [kibbu.md](https://kibbu.io/kibbu.md): the whole site in one Markdown file, written for agents, with the claims, the cost formula, what each number rests on, and when not to use kibbu
- [Quickstart](https://kibbu.io/quickstart): point an OpenAI client at kibbu - the three fields that change, in TypeScript, Python and curl

## Pages

- [Home](https://kibbu.io/): what kibbu does, the hardware estimator and the common questions
- [Solutions](https://kibbu.io/solutions): how the routing works - install, route locally first, fall back to cloud
- [Use cases](https://kibbu.io/use-cases): agents, evals, embeddings, synthetic data and batch processing, with a first pilot and what to measure for each, plus Dify, Databricks, n8n and Langflow setup
- [Company](https://kibbu.io/company): who builds kibbu and why
- [Knowledge hub](https://kibbu.io/blog): guides on AI inference cost

## Comparisons

- [kibbu vs Ollama on one machine](https://kibbu.io/compare/ollama): one developer's machine versus a company fleet behind one API
- [kibbu vs an AI gateway](https://kibbu.io/compare/ai-gateway): routing between other people's clouds versus routing to hardware you already own

## Optional

- [Sign up / console](https://app.kibbu.io): get an API key
- [Privacy](https://kibbu.io/privacy)
- [Terms](https://kibbu.io/terms)