Scalaris Atlas

What Scalaris Atlas is

Scalaris Atlas turns a computer into a local host for AI models, serving them to compatible clients. It accepts OpenAI-compatible clients at localhost and offers routes for chat completions, embeddings, and model lists. Atlas includes Pulsar, and its model manager lets users install other models. The service runs on macOS, Linux, and Windows, and the maker says models are served locally without data leaving the machine. Egress is denied by default, and Atlas refuses non-loopback network bindings unless the user changes that setting. The maker also says it has no telemetry and does not phone home. Scalaris Toolbox handles installation and supervision. Configuration is kept in one TOML file, with settings reloaded when the file changes. Atlas implements only a strict subset of the OpenAI HTTP surface; unimplemented endpoints return a typed 501 response. If GPU memory is exhausted, it falls back to CPU only when `--allow-cpu` is enabled. Atlas has a free plan and API access.

Who it suits

Atlas suits people who use OpenAI-compatible clients and want to run models locally on macOS, Linux, or Windows.

Limits

The API supports a strict subset of the OpenAI HTTP surface, and unimplemented endpoints return a typed 501 response. CPU fallback is available only when `--allow-cpu` is enabled.

Getting started

1. Visit the Scalaris website.
2. Install Atlas through Scalaris Toolbox.
3. Use Atlas’s model manager to install models as needed.
4. Point an OpenAI-compatible client to localhost:8787.
5. Enable `--allow-cpu` if you want CPU fallback when GPU memory is exhausted.

Verdict

Scalaris Atlas offers local model serving through an OpenAI-compatible API, with network restrictions enabled by default. Its limited API surface and conditional CPU fallback matter when choosing it for a client or machine.