BookFreqSelf-hosted AI infrastructure

Private infrastructure · Public API endpoint

AI inference,
run on my own hardware.

BookFreq serves an authenticated, OpenAI-compatible API backed by vLLM on Kubernetes. The inference, networking, and infrastructure are built and operated in my homelab — not rented from an AI API provider.

Current primary model: Qwen3-14B-AWQ · NVIDIA RTX 5060 Ti · K3s

The API

An OpenAI-compatible interface

The direct inference endpoint uses HTTP Basic authentication. These standard API routes support model discovery, chat completions, and text completions; documentation and service health are available separately.

Available models

/v1/models

Discover models served at the direct inference endpoint.

Chat completions

/v1/chat/completions

Send messages using the familiar OpenAI chat schema.

Text completions

/v1/completions

Generate text through the completions API.

Behind the endpoint

There's a whole platform behind this API.

I built it to learn AI infrastructure by actually operating it: a seven-node K3s cluster, two GPU inference tiers, retrieval-augmented generation, GitOps, observability, and a daily human-reviewed vulnerability workflow. The project page has the architecture, design choices, and recorded demonstrations.

See the project