Get started
Atlas is one binary that scales from a laptop to a fleet. Every path ends the same way: an
ANTHROPIC_BASE_URL (and an OpenAI-compatible base URL) you point existing tools at.
Two steps
Section titled “Two steps”- Install Atlas — Homebrew, a one-line script, or the container image.
- Quickstart — serve a model with
atlas upand drive Claude Code against it.
Which path fits you
Section titled “Which path fits you”| You have… | Engine | Path |
|---|---|---|
| A laptop / dev box (no GPU) | llama.cpp | Quickstart |
| One rented cloud GPU | vLLM | Deploy → single GPU box |
| Several machines, one endpoint | either | Deploy → cloud fleet |
A small model on a laptop is great for development, evals, and offline work, but drives Claude Code only intermittently. For reliable agentic use, serve a capable model on a GPU.