Run open models on your own machine. Docker-first, no API keys.

A working, pinned Docker Compose stack - not a list of links. Bring it up,
verify it, and reach your first completion.

WHAT YOU GET

• docker-compose.yml wiring Ollama, llama.cpp and Hugging Face TGI
• .env.example with pinned image tags for reproducible deploys
• README quickstart that gets you to a first local completion
• verify.sh - checks every service is actually serving, not just running
• RUNBOOK.md - operations, troubleshooting, backups, upgrades
• fetch-sources.sh - pulls each project at an exact pinned commit
• docs/ - one page per component explaining its role and when to use it

WHY THESE THREE

Ollama gives you an OpenAI-compatible endpoint with one-command model pulls.
llama.cpp is the low-level GGUF engine when you want control over quantisation
and context. TGI adds continuous batching and token streaming for production
serving. Together they cover local development through to a real endpoint.

ALL OPEN SOURCE, ALL PERMISSIVE

Every component is MIT or Apache-2.0. Full attribution with pinned commits is
included. You are buying the integration layer and operational knowledge - the
source is freely available upstream, and the bundle tells you where and how.

REQUIREMENTS

Docker Engine 24+ with Compose v2. 16 GB RAM for a 7B model on CPU, or an
NVIDIA GPU with the container toolkit for real throughput.

Instant delivery. One-time purchase, free updates.